Enhancing Biometric Authentication through Multimodal Approach Combining Face and Fingerprint Recognition Using Convolutional Neural Networks (CNN)

Enhancing Biometric Authentication through Multimodal Approach Combining Face and Fingerprint Recognition Using Convolutional Neural Networks (CNN) | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Enhancing Biometric Authentication through Multimodal Approach Combining Face and Fingerprint Recognition Using Convolutional Neural Networks (CNN) Usman Abdul Gimba, Noor Afiza Binti Mohd Ariffin, Nur Izura Binti Udzir, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7048774/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 30 Oct, 2025 Read the published version in Discover Computing → Version 1 posted 10 You are reading this latest preprint version Abstract A novel multimodal biometric authentication system combining face and fingerprint verification to ensure enhanced security, accuracy, and resilience in user identification, is presented in this work. The system utilizes Convolutional Neural Networks (CNNs) for effective feature extraction from both biometric modalities, addressing challenges such as occlusion, lighting, and finger quality in real-world scenarios. The results prove the performance of the system, with unimodal face authentication achieving 99.66% accuracy, unimodal fingerprint authentication reaching 100% accuracy, and the multimodal system is 98.35% accurate overall. The multimodal approach significantly reduces False Rejection Rate (FRR) and False Acceptance Rate (FAR), enhancing authentication security and user convenience. The combination of fingerprint and face modalities allows for improved performance by compensating for the weaknesses of individual modalities. The study highlights the potential of deep learning in biometric systems, providing a robust solution for secure access control in mobile and high-security applications. Future work will focus on the further enhancement of the system, for real-time deployment on mobile devices, expanding the model's applicability across diverse environments, and exploring the integration of additional biometric modalities. Unimodal authentication Multimodal authentication CNN Face recognition Fingerprint Figures Figure 1 1 Introduction Information and Communication Technology (ICT) plays an increasingly vital role in everyday life, facilitating access to a wide range of services through personal devices such as smartphones. These gadgets not only enhance communication but also provide users with convenient access to modern features and applications ​[ 1 ]​. The adoption and usage of mobile devices are growing rapidly, driven by their integration into both personal and professional domains. Given the sensitive nature of the data stored on these devices, ensuring robust security has become a critical concern ​[ 2 ]. As highlighted, effective authentication schemes must strike a balance between usability and security. In today’s digital environment, the demand for reliable and secure person authentication systems is more pressing than ever. Conventional authentication schemes such as passwords, PINs, tokens, or ID cards are increasingly viewed as inadequate due to their susceptibility to theft, forgery, and forgetting [ 3 ]. Consequently, biometric-based authentication systems have gained prominence. These systems leverage unique physiological or behavioural traits to verify identity ​[ 4 ]. Over the past few decades, biometric technologies have seen widespread adoption across various domains, largely due to rapid technological advancements. This growth has been further accelerated by the increasing prevalence of digital fraud, which underscores the need for dependable authentication mechanisms to secure transactions and protect access to both digital and physical systems, such as mobile devices and institutional facilities ​[ 5 ]​. Most early biometric systems relied on a single trait such as fingerprint or facial recognition for user authentication. These unimodal systems, while relatively simple to develop, are vulnerable to spoofing attacks and often yield lower recognition accuracy ​[ 6 ]​. To address these limitations, multimodal biometric authentication systems have been proposed. These systems integrate multiple biometric traits and are generally considered more secure and reliable than their unimodal counterparts. Commonly used biometric modalities include electrocardiogram (ECG), fingerprint, voice, iris, and face. Multimodal systems can operate in various modes, including serial, parallel, or hierarchical, depending on the application requirements ​[ 7 ]​. Research indicates that multimodal approaches significantly reduce error rates and offer greater resistance to forgery attempts. Among the various biometric traits, fingerprint and facial recognition remain the widely used because of their user-friendliness and high accuracy. Fingerprint recognition is favoured for its simplicity in acquisition, straightforward processing, and long-term stability ​[ 8 ]. Facial recognition, on the other hand, enables identification based on visual appearance and is frequently used because of its user-acceptability and non-intrusive nature [ 9 ]​. The focus of this study is user authentication for mobile devices, with face recognition and fingerprint verification. In this paper we used haar cascade for face detection and minutiae algorithm for fingerprint detection while for feature extraction convolution neural network is used. The main goal of this study is to significantly prevent unauthorized identification and impersonation, with the goal of improving the accuracy for user authentication. Two publicly available datasets were used for face recognition and fingerprint verification, Georgia Tech face database (gt_db), Essex dataset, FVC2000 and SOCOFing dataset. Preprocessing the datasets reduces the effect of noise interference and occlusion on the detection results. Secondly, convolutional neural networks are used to do feature extraction for matching and classification. Finally, an experimental validation is conducted utilizing the proposed technique to assess its accuracy and robustness on Georgia Tech face database (gt_db), Essex dataset, FVC2000 and SOCOFing dataset to experiment on. The main contributions of this study are as follows : The various existing research on biometric authentication both unimodal and multimodal, and challenges faced on accurate authentication are highlighted We developed an Improved hybrid algorithm using machine learning and deep learning, the suggested algorithm achieved a high Accuracy. Comprehensive experimental comparisons were performed with the existing algorithms, an Accuracy of 98.35% was achieved higher than the previous. 2 Related works Venkata Ramana et al. ​[ 10 ]​ developed a multimodal system that integrates biometrics face, fingerprint, iris, palm, and ear using KLDA for feature reduction and deep learning for classification, achieving high accuracy. However, the complexity of simultaneously acquiring multiple biometric inputs could hinder real-world usability on mobile devices. Tahri & Beladgham ​[ 11 ] presents a fuse facial and voice features at the feature level using SincNet and CNN architectures to achieve high identification rates and low error margins, enhancing multimodal security. However, Environmental factors like noise and lighting variability in mobile settings could significantly degrade performance. Amit Kumar et al. [ 12 ]The authors analyze fusion of face and gait biometrics using various ML models, demonstrating superior accuracy over unimodal approaches. However, While the fusion approach is sound, experiments were limited to static datasets. Amber Hayat et al. [ 13 ]​ utilized​ a genetic algorithm to dynamically weight face, fingerprint, and ear biometric traits at the score fusion level, achieving 97% accuracy. However, while promising for robustness, the computational load of the genetic algorithm and its performance in noisy mobile settings are not addressed. Divan & Gulhane ​[ 14 ]presented a two-stage fingerprint matching system for secure voting, using minutiae-based algorithms. It emphasizes privacy protection by combining templates from two different fingerprints to lower FRR rates. However, the work remains conceptual without real-world testing. Riseul Ryu et al. ​[ 15 ] This review synthesizes continuous multimodal authentication strategies, highlighting the dominance of score-level fusion and the need for greater attention to scalability and usability in real-world deployments. However, the review does not propose concrete methods for addressing mobile-specific limitations. Srivastava et al. [ 16 ] suggested a real-time face verification system that fuses Gabor Wavelet Transform (GWT) with Scale-Invariant Feature Transform (SIFT) and Speeded-Up Robust Features (SURF) to achieve better recognition performance and computational efficiency. The proposed method involves facial image decomposition into GWT-based sub-bands, SIFT or SURF feature extraction, and score-level fusion for verification. Benchmarked on PUT and ORL databases, the proposed GWT-SURF model achieved 98.75% accuracy in 3.4 seconds of run-time for every 100 images, outperforming conventional PCA, LDA, and basic SIFT/SURF approaches. The system was also determined to be robust against changes in lighting and image entropy, rendering it suitable for real-time biometric system. However, no comparative evaluation with modern deep learning systems and real-deployment test. Chenhao Lin et al. ​[ 17 ] presented​ CrossBehaAuth, a deep learning-based system for cross-device keystroke authentication, leveraging CNN-GRU architecture and data augmentation to address session and device variation. However, the work is narrowly focused on keystroke dynamics and lacks integration with broader multimodal authentication frameworks. Ammour et al. ​[ 18 ] devised a novel multimodal biometric system integrates Electrocardiogram (ECG) signals and fingerprint to counteract spoofing attacks. Using a transformer-based deep learning model, it demonstrates robust security enhancements. Jhansi Bharathi Madavarapu et al. ​[ 19 ] study investigates machine learning usage, particularly LSTM models, for behavioral biometrics like keystroke dynamics and mouse movement, achieving high accuracy. However, although effective, the study centres on desktop environments only. Swati Singh et al. ​[ 20 ] presented a broad overview of biometric technologies and the impact of artificial intelligence on fingerprint, facial, iris, and voice recognition methods. However, the survey lacks detailed exploration of multimodal fusion strategies or practical security system designs. Noor Afiza et al. [ 21 ]​ This study presents a novel system for face detection that combines the Haar cascade technique with Convolutional Neural Networks (CNNs). Experimental results highlight the strong performance of this hybrid model, showing a notable improvement in accuracy compared to conventional machine learning methods. The combined use of Haar cascades and CNNs proves to be more effective in detecting faces than traditional approaches. However, despite the promising accuracy, the research lacks a thorough evaluation of key performance metrics such as precision and the F1 score, which are crucial for assessing the overall robustness and dependability of the system. Zheng Hui Goh et al. [ 22 ]​ propose a multimodal authentication framework using IoM hashing and Alignment-Free Hashing (AFH) to secure templates, fusing face, fingerprint, iris, and finger-vein biometrics. The system offers strong security properties irreversibility, revocability, unlinkability. However, despite outstanding theoretical robustness, its scalability to resource-constrained mobile environments remains untested, posing challenges for real-world application. Yusuf Magaji et al. ​[ 23 ] proposed authentication system combines fingerprint and facial recognition to verify the identity of mobile device users. Access to the device is granted only if both authentication methods are successfully completed. However, facial recognition relies on connectivity to Microsoft servers to function. Haq et al. ​[ 24 ] evaluated various detection and recognition methods, for face recognition and face detection and compared them based on parameters like accuracy, speed and computational complexity. Strength and limitations of the methods were discussed in the review paper, with future research direction focused improving algorithm robustness, developing lightweight models for mobile devices and addressing dataset biases to enhance fairness and efficiency in face recognition. However, no clear suggestion on how to improve the alogrithms. 3 Methodology The proposed multimodal biometric authentication system integrates facial and fingerprint modalities to enhance identity verification accuracy and robustness, especially in scenarios where unimodal system fails due to occlusion, spoofing and environmental noise. The methodological framework consists of four key stages: data pre-processing, feature extraction via Convolution Neural Network (CNN), multimodal fusion, and classification. Figure 1 illustrates the proposed system’s flowchart The figure illustrates the overall process of how the proposed multimodal biometric authentication. I) sensors layer where is the use will input both this face and fingerprint. II) preprocessing layer ensures the raw data were properly normalized and ROI features are extracted. III) feature extraction layer, identifies and isolates relevant visual patterns. IV) Matching layer, used to compared the inputted biometrics with the biometrics stored in the database. V) fusion score layer, the metrics used to evaluate the quality of the image inputted. Once the fusion score reached a high level the user is granted access to the device “Authenticate”. 3.1 Pre-processing Pre-processing is a modality-specific to ensure that raw data were properly normalized and that region-of-interest (ROI) features were clearly extracted. This involved several steps, such as cleaning, transforming, and normalizing the data, all aimed at enhancing the accuracy and performance of the model. 3.1.1 Face Detection using Haar Cascade The Haar Cascade method is a widely adopted approach for object detection within images, regardless of variations in scale or position. Introduced by Viola and Jones, this real-time and computationally efficient algorithm utilizes Haar feature-based cascade classifiers. The technique relies on machine learning principles, wherein a cascade function is trained using a substantial database comprising both negative and positive image samples. Haar features effectively capture grayscale variations in facial images, as they are derived from the differences in pixel intensity values, as demonstrated by Zhang et al. [ 25 ] ​ Facial images were processed using the (haarcascade_frontalface_default.xml) Haar Cascade classifier from OpenCV based on the Viola-Jones algorithm. Each RGB image was first converted to grayscale for computational efficiency, and facial regions were detected using the detectMultiScale method. The first detected face (if any) was extracted as the region of interest (ROI), resized to 128×128 pixels, and normalized to intensity values between 0 and 1. This method enables consistent face localization under varying lighting and pose conditions, providing a clean input for the CNN. This method effectively standardizes the facial input, mitigating challenges posed by background noise, varying illumination, and scale differences. 3.1.2 Fingerprint Minutiae Extraction A minutiae-based fingerprint recognition algorithm emphasizing local feature extraction. Initial preprocessing includes intensity normalization based on a global threshold, followed by binarization using a NOT operation to invert grayscale values, ensuring ridge pixels are represented as 1s and furrows as 0s. Morphological OPEN and CLOSE operations are applied to extract the Region of Interest (ROI) and suppress background noise, while ridge thinning reduces ridge structures to a single-pixel width without altering topology. Minutiae points are extracted using a 3×3 window, identifying bifurcations when the center pixel with value 1 is adjacent to exactly three neighboring 1s. To mitigate false detections, spatial redundancy is reduced by eliminating bifurcation points within one-pixel proximity based on four-neighbor analysis. False minutiae are eliminated during post-processing by computing the average inter-ridge distance D and comparing it with the distance d between minutiae pairs. If d < D and both points lie on the same ridge, they are considered spurious and removed final fingerprint recognition ​[ 26 ] Fingerprint samples were preprocessed through grayscale A global thresholding operation (cv2.THRESH_BINARY_INV) was used to segment foreground ridge structures. The binarized fingerprint was then skeletonized using morphological thinning (skimage.morphology.skeletonize) to reduce ridges to a single-pixel width for accurate feature localization.Minutiae points including ridge endings and bifurcations were extracted using the Harris corner detector and corner_peaks. The detected minutiae were rendered onto a blank canvas to emphasize discriminatory structures. All images were resized to 128×128 pixels to match the face input dimensions. This process not only improves the signal-to-noise ratio but also converts complex biometric patterns into simplified, high-information-density formats suitable for CNN-based learning. 3.2 Feature Extraction using Convolutional Neural Networks (CNN) Convolutional Neural Networks (CNNs) have significantly transformed the fields of image analysis and computer vision by emulating the functionality of the human visual system to recognize complex visual features. These networks are composed of several key layers: convolutional layers filter-based for extracting patterns from input data, pooling layers that compress the spatial dimensions of the resulting feature maps, and activation functions that incorporate non-linear transformations. Through training methods such as backpropagation and gradient descent, CNNs are capable of automatically learning the most effective filters from the data itself, thereby eliminating the need for manual feature engineering. The versatility has enabled breakthroughs in fields such as object recognition, video analysis, and medical imaging. In this study, facial and fingerprint images were integrated into the CNN model due to its robust capabilities in handling image-based data [ 27 ]​. Deep hierarchical characteristics are extracted from both modalities, using a dual-branch CNN architecture. Each branch independently processes one modality using shared architectural design principles. 3.2.1 CNN Architecture Each input (face or fingerprint) passed through the following layers: Input Layer : Accepts a tensor of shape (128, 128, 1). Convolutional Layers : Three blocks of 2D convolution with ReLU activation and same padding were used to capture edge, texture, and region-based features. Each block contained 32, 64, and 128 filters, respectively. Batch Normalization : Applied after each convolutional layer to stabilize learning and improve convergence. Max Pooling : Reduces spatial dimensions while preserving dominant features. Flatten + Dense Layers : The final convolutional outputs were flattened and passed through a fully connected layer of 512-neuron with ReLU activation and dropout regularization (0.5) to avoid overfitting. Each branch independently learned modality-specific representations while maintaining a consistent network depth and filter sizes for compatibility. 3.2.2 Feature Fusion and Multimodal Classification The flattened outputs of both CNN branches were concatenated to form a unified feature vector. This early fusion approach allowed the model to learn cross-modal dependencies. The fused vector passed through a shared 512-neuron fully connected layer, followed by a dropout regularization (0.5) to mitigate overfitting and softmax output layer for final classification. The Adam optimizer with categorical cross-entropy loss was used to train the system and evaluated on a hold-out test set. 3.3 Dataset and Implementation All face and fingerprint images were matched by subject IDs to form multimodal pairs. Both datasets were pre-processed and split using an 80% for training and 20% for testing. TensorFlow was used to train the model for 20 epochs with 32 batch size, with automatic prefetching and batching via tf.data.Dataset. A checkpointing mechanism ensured the best-performing model on the validation set was saved. The final model was evaluated using categorical accuracy and saved in .keras format. These datasets ensure diversity in biometric characteristics and are suitable for evaluating a real-world authentication system. Face Dataset : Georgia Tech face database (gt_db), Essex face database was used, with subject-wise folders for label annotation, images are stored in a directory structure for supervised learning. Fingerprint Dataset : FVC2000 and SOCOFing dataset was used, containing real and altered fingerprint images across multiple classes, categorized by subject and alteration types 3.3.1 Essex face database [ 28 ] The public datasets utilized in this study have also been employed in prior researchers. The Faces94 dataset includes 180x200 pixel images of 153 individuals, featuring a green background, varied head orientations, minimal facial expression changes, and a range of poses. The Faces95 dataset comprises images of 72 individuals, also sized at 180x200 pixels, with frontal views and a red background created using flames. In the Faces96 dataset, there are 152 individual's images, each with a resolution of 196x196 pixels. Additionally, the Grimace dataset has 18 individual's images at 180x200 pixel resolution. To standardize input dimensions and reduce file size while preserving image quality, all images were resized to 128x128 pixels. 3.3.2 SOCOFing dataset​[ 29 ] The SOCOFing dataset comprises six thousand fingerprint images from six hundred African individuals, each contributing ten fingerprints. All participants are aged 18 and above. The dataset is enriched with metadata, including labels for gender, hand orientation, and specific finger identification. In addition to the original images, synthetically modified versions are included, containing z-cut, central rotation, and three different levels of modification obliteration produced with the STRANGE toolkit. The original fingerprint impressions were captured using Hamster Plus and SecuGen fingerprint scanners. Each image is formatted in grayscale with a resolution of 1 × 96 × 103 (gray × width × height). 3.3.3 Georgia Tech face database (Gt_Db)​ [ 30 ] The Georgia Tech face database (Gt_Db) comprises 50 classes, each containing 15 face images. These images were captured at various times, without a specific lighting setup, and feature complex backgrounds. The data collection was conducted in stages over approximately five months. The sample face images in the Gt_Db are sized at 640×480. 3.3.3 FVC2000 [ 31 ]​ FVC2000, the "universal" sensor comprises four different sensors to reflect advancements in fingerprint sensing. Small, inexpensive optical and capacitive sensors were used to collect data for Databases 1 and 2, respectively. Database 3 used a large-area optical sensor with a high quality whereas, Database 4 included artificially created images. For algorithm tuning, each database had 880 fingerprints from 110 fingers, divided into a training-set of 80 photos and a test-set of 800 images, set B and set A respectively. 4. Result This section explains the result obtained in this study, both unimodal and multimodal comparison are presented. The multimodal biometric authentication system was developed using a Convolutional Neural Network (CNN). The implementation was carried out in Python using the OpenCV open-source library. Key performance indicators like accuracy, False Acceptance Rate (FAR), and False Rejection Rate (FRR) are used to assess the system's effectiveness. Experimental testing was conducted using both unimodal and multimodal authentication approaches across various datasets. Table 1 Unimodal Face authentication Authentication Method Accuracy (%) FRR (%) FAR (%) Face Authentication (Georgia Tech face database) Gt_db ​​[ 25 ]98.78 1.22 0.98 ​​[ 32 ]​94.29 - - [Our method]99.66 0.339 0.001 The performance of unimodal face authentication presented in Table 1 above, demonstrates the effectiveness of the face recognition system. The system demonstrated a high accuracy rate of 99.66%, accompanied by 0.34% of False Rejection Rate (FRR) and an exceptionally 0.001% low False Acceptance Rate (FAR). These results highlight how well the proposed face authentication method performs, surpassing earlier approaches in both accuracy and error rates. The low FAR is particularly important for ensuring that unauthorized users are not mistakenly accepted, while the low FRR indicates that legitimate users are unlikely to be rejected. Table 2 Unimodal Fingerprint authentication Authentication Method Accuracy (%) FRR (%) FAR (%) Fingerprint (Sokoto fingerprint dataset) SOCOFing ​​[ 33 ]99.78 - - ​​[ 34 ]​97.47 - - [Our method]100.00 0.00 0.00 The above Table 2 , displays the performance of unimodal fingerprint authentication using the SOCOFing dataset, which achieved 100% accuracy, with zero FRR and zero FAR. This result underscores the robustness and reliability of fingerprint recognition, making it an excellent candidate for secure authentication. The perfect accuracy and zero error rates validate the effectiveness of the minutiae-based and CNN algorithm used for fingerprint matching, ensuring both high security and user convenience. Despite the high performance of both unimodal systems, challenges such as vulnerability to spoofing attacks (e.g., fake fingerprints or facial images) and environmental factors (e.g., lighting for face recognition, skin quality for fingerprint matching) persist. These drawbacks emphasize the necessity of a more thorough strategy that incorporates both modalities. Table 2 Multimodal Biometric Authentication Database Authentication Method Accuracy (%) FRR (%) FAR (%) Gt_db and TIMIT Ref​[ 25 ] 90.28 9.72 9.29 Labelled Faces in the Wild (LFW) and FVC2000 Ref​[ 35 ] 96.26 3.64 0.0019 gt_db and FVC2000 Our Method 91.12 8.88 0.39 Essex face and SOCOFing Our Method 98.35 1.658 0.01 When combining both face recognition and fingerprint verification, the multimodal biometric authentication system showed remarkable improvement in performance, as presented in Table 3 above. The multimodal system attained 98.35% accuracy, with 1.658% FRR and 0.01% FAR. These results indicate that multimodal fusion enhances the overall accuracy, ensures a very low False Acceptance Rate (FAR), and reduces False Rejection Rates (FRR). The low FAR is especially significant, as it indicates the system's high resistance to false acceptances, which is essential for securing sensitive applications such as mobile devices and access control systems. The multimodal approach benefits from the complementary strengths of face recognition and fingerprint verification. Face recognition, while non-intrusive, can be affected by poor lighting or occlusions, while fingerprint recognition may struggle with finger placement issues. By combining these modalities, the multimodal system compensates for the weaknesses of each individual modality, leading to a more robust and secure authentication solution. Conclusion In conclusion, a novel multimodal biometric authentication was introduced by combining face recognition and fingerprint verification, while for feature extraction Convolutional Neural Networks (CNNs) was used. The integration of these two biometric modalities proved effective in improving system’s reliability, security, and accuracy. Experimental findings revealed that the face-only authentication achieved a notable accuracy of 99.66%, accompanied by minimal False Rejection Rate (FRR) and False Acceptance Rate (FAR). Meanwhile, fingerprint-authentication only attained a perfect accuracy of 100%, establishing a strong performance standard for fingerprint-based authentication systems. The multimodal system further enhanced these results, achieving 98.35% accuracy, with significantly lower FRR (1.658%) and FAR (0.01%) compared to traditional unimodal methods. This improvement demonstrates the effectiveness of modality fusion, where the strengths of face recognition and fingerprint verification complement each other, compensating for their respective weaknesses in real-world conditions such as occlusion, lighting variability, and finger quality issues. The use of CNNs in feature extraction was critical to the system's success, enabling the automatic learning of optimal features from raw biometric data without requiring manual intervention. This approach not only improved the model's generalization ability but also enhanced its performance in dynamic environments. Despite the strong results, challenges such as computational complexity and the quality of input data remain. Further research is needed to optimize the system for processing in real time on mobile devices and resource-constrained environments. Additionally, integrating behavioral biometrics could further strengthen the system's security and adaptability. In summary, the study's findings highlight how successful multimodal biometric systems in providing secure and dependable authentication across diverse applications, such as mobile devices. By harnessing deep learning techniques and combining multiple biometric modalities, this research provides a strong basis for future developments in biometric security technologies. Declarations Acknowledgement: This study is supported by Petroleum Trust Development Fund (PTDF) Nigeria, Universiti Putra Malaysia. We are really grateful for the facilities and funding provided, which made the publication of this study possible. Author Contributions: Usman Abdul Gimba: Manuscript writing, methodology and data analysis, Noor Afiza Binti Muhd Ariffin: Manuscript reviewing, supervision, Nur Izura Udzir: Manuscript reviewing, supervision, Nor Fazlida Mohd Sani: Manuscript reviewing, supervision. All authors reviewed the manuscript and approved the final version of the manuscript. Availability of Data and Materials: The data used for this study can be obtained from the corresponding author upon request. Ethics Approval: Not applicable Conflicts of Interest: No conflict of interest regarding this study shown by the authors Clinical Trial Number: Not applicable Consent to Participate: Not applicable. Consent to Publish: Not applicable. References Gimba, U. A., & Ariffin, N. A. M. (2025). Review on User Authentication on Mobile Devices. Journal of Advanced Research in Applied Sciences and Engineering Technology , 46 (2), 26–36. Olade, I., Liang, H., ning, Fleming, C.. A Review of Multimodal Facial Biometric Authentication Methods in Mobile Devices and Their Application in Head Mounted Displays. In:, SmartWorld, I. E. E. E., Intelligence, U., Computing, Advanced, Computing, T., Computing, S., Communications, & Cloud (2018). & Big Data Computing, Internet of People and Smart City Innovation (SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI) [Internet]. IEEE; 2018. pp. 1997–2004. Available from: https://ieeexplore.ieee.org/document/8560312/ El-Rahiem, B. A., El-Samie, F. E. A., & Amin, M. (2022). Multimodal biometric authentication based on deep fusion of electrocardiogram (ECG) and finger vein. Multimedia Systems (pp. 1325–1337). Springer Science and Business Media Deutschland GmbH. Sengar, S., Singh, U., Hariharan, K., & Rajkumar (2020). International Conference on Emerging Smart Computing and Informatics (ESCI): AISSMS Institute of Information Technology, Pune, India. Mar 12–14, 2020. IEEE; 2020. 309–312 p. Alshardan, A., Kumar, A., Alghamdi, M., Maashi, M., Alahmari, S., Alharbi, A. A. K. (2024). Multimodal biometric identification: leveraging convolutional neural network (CNN) architectures and fusion techniques with fingerprint and finger vein data. PeerJ Comput Sci . ;10. Dargan, S., & Kumar, M. (2020). A comprehensive survey on the biometric recognition systems based on physiological and behavioral modalities (Vol. 143). Elsevier Ltd. Expert Systems with Applications. Sajja, S. L., Mukesh, S., Hussein, A. H. A., Sunil, G., & Habelalmateen, M. I. (2023). Multimodal Biometric Authentication System using Probabilistic Fuzzy based Tuna Search Optimization. In: IEEE 1st International Conference on Ambient Intelligence, Knowledge Informatics and Industrial Electronics, AIKIIE 2023. Institute of Electrical and Electronics Engineers Inc. Madduluri, S., & Kumar, T. K. (2024). Multimodal Biometric Authentication System using ECG Signals and Fingerprints. In: 4th International Conference on Sustainable Expert Systems, ICSES 2024 - Proceedings. Institute of Electrical and Electronics Engineers Inc.; pp. 115–21. Yadav, A. K., Pateriya, R. K., Gupta, N. K., Gupta, P., Saini, D. K., Alahmadi, M., & Computers (2022). Materials and Continua . ;72(2):2697–2712. Venkata Ramana, S., AN ENHANCED MULTI-MODAL BIOMETRIC AUTHENTICATION SYSTEM, S. V. R., & USING MODIFIED DEEP LEARNING MODEL. (2023). Journal of Science and Technology . ;8(12):147–115. Merit, K., & Beladgham, M. (2024). Enhancing Biometric Security with Bimodal Deep Learning and Feature-level Fusion of Facial and Voice Data. Journal of Telecommunications and Information Technology . ;(4):31–42. Kumar, A., Jain, S., & Kumar, M. (2024). Comparative Study of Multi-Biometrics Authentication Using Machine Learning Algorithms. In: 2024 11th International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions), ICRITO 2024. Institute of Electrical and Electronics Engineers Inc. Hayat, A., Kumar, A., Kumarbhateja, A., Pal, S. K.. An Approach for Multimodal Biometric Authentication using Genetic Algorithm. In: 15th International Conference on Computing Communication and, & Technologies, N. (2024). ICCCNT 2024. Institute of Electrical and Electronics Engineers Inc.; 2024. Divan, T. A Fingerprint Matching Technique using Minutiae based Algorithm for Voting System: A Survey. Ryu, R., Yeom, S., Kim, S. H., & Herbert, D. (2021). Continuous Multimodal Biometric Authentication Schemes: A Systematic Review (Vol. 9, pp. 34541–34557). IEEE Access. Institute of Electrical and Electronics Engineers Inc. Srivastava, R., Tomar, R., Sharma, A., Dhiman, G., Chilamkurti, N., & Kim, B. G. (2021). Real-time multimodal biometric authentication of human using face feature analysis. Computers Materials and Continua . ;69(1). Lin, C., He, J., Shen, C., Li, Q., & Wang, Q. (2023). CrossBehaAuth: Cross-Scenario Behavioral Biometrics Authentication Using Keystroke Dynamics. IEEE Trans Dependable Secure Comput , 20 (3), 2314–2327. Ammour, N., Bazi, Y., & Alajlan, N. (2023). Multimodal Approach for Enhancing Biometric Authentication. J Imaging . ;9(9). Madavarapu, J. B., Mittal, M., Salagrama, S., Adnan, M. M., Rana, A., & Yadav, K. (2024). Behavioral Biometrics Authentication Systems: Leveraging Machine Learning for Enhanced Cybersecurity. In: Proceedings of International Conference on Communication, Computer Sciences and Engineering, IC3SE 2024. Institute of Electrical and Electronics Engineers Inc.; pp. 1478–83. Singh, S., Sharma, S., Awasthi, M., Rawat, S., & Chanti, Y. (2024). Advancements of Emerging Technologies in Biometrics Authentication. In: 2024 IEEE 1st Karachi Section Humanitarian Technology Conference, Khi-HTC 2024. Institute of Electrical and Electronics Engineers Inc. Mohd Ariffin, N. A., Gimba, U. A., & Musa, A. (2025). Face Detection based on Haar Cascade and Convolution Neural Network (CNN). Journal of Advanced Research in Computing and Applications [Internet]. ;38(1):1–11. Available from: https://www.akademiabaru.com/submit/index.php/arca/article/view/5576 Goh, Z. H., Wang, Y., Leng, L., Liang, S. N., Jin, Z., Lai, Y. L., et al. (2022). A Framework for Multimodal Biometric Authentication Systems With Template Protection. Ieee Access : Practical Innovations, Open Solutions , 10 , 96388–96402. Yusuf, M., Abdul Gimba, U., Uthman Bello, A., Habu Adamu, A., & Salisu, S. (2019). Two Way Authentication for Android Mobile Phones. Vol. 5, Dutse Journal of Pure and Applied Sciences (DUJOPAS). Jun. Ul Haq, M., Sethi, M. A. J., Ahmad, S., Ahmad, N., Anwar, M. S., & Kutlimuratov, A. (2025). A Comprehensive Review of Face Detection/Recognition Algorithms and Competitive Datasets to Optimize Machine Vision (Vol. 84, pp. 1–24). Tech Science. Computers, Materials and Continua. Zhang, X., Cheng, D., Jia, P., Dai, Y., & Xu, X. (2020). An Efficient Android-Based Multimodal Biometric Authentication System with Face and Voice. Ieee Access : Practical Innovations, Open Solutions , 8 , 102757–102772. Narwal, S., & Kaur, D. (2016). Comparison between Minutiae Based and Pattern Based Algorithm of Fingerprint Image. International Journal of Information Engineering and Electronic Business [Internet]. ;8(2):23–9. Available from: http://www.mecs-press.org/ijieeb/ijieeb-v8-n2/v8n2-3.html Salturk, S., & Kahraman, N. (2024). Deep learning-powered multimodal biometric authentication: integrating dynamic signatures and facial data for enhanced online security. Neural Computing And Applications , 36 (19), 11311–11322. Aminu, M., & Ahmad, N. A. (2022). Locality preserving partial least squares discriminant analysis for face recognition. Journal of King Saud University - Computer and Information Sciences , 34 (2), 153–164. Isah Shehu, Y., Ruiz-Garcia, A., Palade, V., & James, A. Sokoto Coventry Fingerprint Dataset [Internet]. Available from: https://www.kaggle.com/ çoskun, M., Uçar, A., Yildirim, Ö., Demir, Y.. MEES 2017: proceedings of the International Conference on Modern Electrical and Energy SystemsKremenchuk Mykhailo Ostrohradskyi National University, Ukraine, 15–17, & November (2017). IEEE; 2018. 376–379 p. Maio, D., Maltoni, D., Cappelli, R., Wayman, J. L., & Jain, A. K. (2000). FVC: Fingerprint Verification Competition. Hamou, A. O., & Chelali, F. Z. (2024). Facial Recognition Based on Machine and Deep Learning. In: ICAEE 2024 - International Conference on Advanced Electrical Engineering 2024. Institute of Electrical and Electronics Engineers Inc. Singh, R., Sharma, N., Chauhan, R., Choudhary, A., & Gupta, R. (2023). Enhanced Fingerprint Alteration Detection Using Lightweight CNN Model Trained on SOCOFing Dataset. In: 2023 3rd International Conference on Smart Generation Computing, Communication and Networking, SMART GENCON 2023. Institute of Electrical and Electronics Engineers Inc. Narayanan, A., & Hameeduddin, Q. M. (2023). Gender Detection and Classification from Fingerprints Using Convolutional Neural Network. In: ICSPC 2023–4th International Conference on Signal Processing and Communication. Institute of Electrical and Electronics Engineers Inc.; pp. 178–81. Jena, P. P., Kattigenahally, K. N., Nikitha, S., Sarda, S., & Harshalatha, Y. (2021). Multimodal Biometric Authentication: Deep Learning Approach. In: 2021 International Conference on Circuits, Controls and Communications, CCUBE 2021. Institute of Electrical and Electronics Engineers Inc. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 30 Oct, 2025 Read the published version in Discover Computing → Version 1 posted Editorial decision: Revision requested 02 Sep, 2025 Reviews received at journal 29 Aug, 2025 Reviews received at journal 20 Aug, 2025 Reviewers agreed at journal 17 Aug, 2025 Reviewers agreed at journal 14 Aug, 2025 Reviewers invited by journal 14 Aug, 2025 Editor assigned by journal 14 Aug, 2025 Editor invited by journal 07 Aug, 2025 Submission checks completed at journal 02 Aug, 2025 First submitted to journal 02 Aug, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7048774","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":502933723,"identity":"4be58209-7787-4508-aee9-1d83a67eb1db","order_by":0,"name":"Usman Abdul Gimba","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8klEQVRIiWNgGAWjYBADHgYJBsYHDAwHYAIJRGlhNiBJCwNQC5sEUVoMzp9O/FzAUCujO7v5WTVPzR05fgbmhw8Y29Jwa7mRu1l6BsNxHrM7x8xu8xx7ZizZwGZswNiWg0cL7wZpHoZjPGY3EoBa2A4nbjjAwybB2FaBx2FnN/+GaEn/VszzjxgtB3K3AW2pAWrJMWPmbYNrwe0wyRu526x5GA4A/XKmWHJu32FjyWagXxLO4fY+H9Bht3kY6uzNbrdv/PDm22E5fvbmhw8+lCXj1AIGjP8Og2kmHhDJzEBURNZBtP4grHIUjIJRMApGIAAAd5lUVm7ulG4AAAAASUVORK5CYII=","orcid":"","institution":"University Putra Malaysia","correspondingAuthor":true,"prefix":"","firstName":"Usman","middleName":"Abdul","lastName":"Gimba","suffix":""},{"id":502933724,"identity":"68ca9f28-8835-452f-a00e-8a92ffb7f0ae","order_by":1,"name":"Noor Afiza Binti Mohd Ariffin","email":"","orcid":"","institution":"University Putra Malaysia","correspondingAuthor":false,"prefix":"","firstName":"Noor","middleName":"Afiza Binti Mohd","lastName":"Ariffin","suffix":""},{"id":502933725,"identity":"106ba2c2-dd77-43f0-9e6e-c9287d697e00","order_by":2,"name":"Nur Izura Binti Udzir","email":"","orcid":"","institution":"University Putra Malaysia","correspondingAuthor":false,"prefix":"","firstName":"Nur","middleName":"Izura Binti","lastName":"Udzir","suffix":""},{"id":502933726,"identity":"34289d3f-5bf6-463e-bab8-33e777b3f3c1","order_by":3,"name":"Nor Fazlida Mohd Sani","email":"","orcid":"","institution":"University Putra Malaysia","correspondingAuthor":false,"prefix":"","firstName":"Nor","middleName":"Fazlida Mohd","lastName":"Sani","suffix":""}],"badges":[],"createdAt":"2025-07-04 17:23:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7048774/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7048774/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s10791-025-09775-z","type":"published","date":"2025-10-30T15:58:54+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":89658728,"identity":"2a68561c-14c3-4da2-ad6b-7b72b33f4458","added_by":"auto","created_at":"2025-08-22 10:45:30","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":83385,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart\u003cstrong\u003e \u003c/strong\u003efor multimodal biometric authentication\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7048774/v1/419fbf1bf8657e27953188b3.png"},{"id":95040593,"identity":"bbe06700-ec32-40f6-8b71-27e211af2d6a","added_by":"auto","created_at":"2025-11-03 16:09:57","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":960866,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7048774/v1/70aae9f9-f186-4031-8b79-dcd1fb2f12bf.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Enhancing Biometric Authentication through Multimodal Approach Combining Face and Fingerprint Recognition Using Convolutional Neural Networks (CNN)","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eInformation and Communication Technology (ICT) plays an increasingly vital role in everyday life, facilitating access to a wide range of services through personal devices such as smartphones. These gadgets not only enhance communication but also provide users with convenient access to modern features and applications ​[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]​. The adoption and usage of mobile devices are growing rapidly, driven by their integration into both personal and professional domains. Given the sensitive nature of the data stored on these devices, ensuring robust security has become a critical concern ​[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eAs highlighted, effective authentication schemes must strike a balance between usability and security. In today\u0026rsquo;s digital environment, the demand for reliable and secure person authentication systems is more pressing than ever. Conventional authentication schemes such as passwords, PINs, tokens, or ID cards are increasingly viewed as inadequate due to their susceptibility to theft, forgery, and forgetting [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Consequently, biometric-based authentication systems have gained prominence. These systems leverage unique physiological or behavioural traits to verify identity ​[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eOver the past few decades, biometric technologies have seen widespread adoption across various domains, largely due to rapid technological advancements. This growth has been further accelerated by the increasing prevalence of digital fraud, which underscores the need for dependable authentication mechanisms to secure transactions and protect access to both digital and physical systems, such as mobile devices and institutional facilities ​[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]​.\u003c/p\u003e\u003cp\u003eMost early biometric systems relied on a single trait such as fingerprint or facial recognition for user authentication. These unimodal systems, while relatively simple to develop, are vulnerable to spoofing attacks and often yield lower recognition accuracy ​[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]​. To address these limitations, multimodal biometric authentication systems have been proposed. These systems integrate multiple biometric traits and are generally considered more secure and reliable than their unimodal counterparts. Commonly used biometric modalities include electrocardiogram (ECG), fingerprint, voice, iris, and face. Multimodal systems can operate in various modes, including serial, parallel, or hierarchical, depending on the application requirements ​[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]​.\u003c/p\u003e\u003cp\u003eResearch indicates that multimodal approaches significantly reduce error rates and offer greater resistance to forgery attempts. Among the various biometric traits, fingerprint and facial recognition remain the widely used because of their user-friendliness and high accuracy. Fingerprint recognition is favoured for its simplicity in acquisition, straightforward processing, and long-term stability ​[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Facial recognition, on the other hand, enables identification based on visual appearance and is frequently used because of its user-acceptability and non-intrusive nature [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]​.\u003c/p\u003e\u003cp\u003eThe focus of this study is user authentication for mobile devices, with face recognition and fingerprint verification. In this paper we used haar cascade for face detection and minutiae algorithm for fingerprint detection while for feature extraction convolution neural network is used. The main goal of this study is to significantly prevent unauthorized identification and impersonation, with the goal of improving the accuracy for user authentication. Two publicly available datasets were used for face recognition and fingerprint verification, Georgia Tech face database (gt_db), Essex dataset, FVC2000 and SOCOFing dataset. Preprocessing the datasets reduces the effect of noise interference and occlusion on the detection results. Secondly, convolutional neural networks are used to do feature extraction for matching and classification. Finally, an experimental validation is conducted utilizing the proposed technique to assess its accuracy and robustness on Georgia Tech face database (gt_db), Essex dataset, FVC2000 and SOCOFing dataset to experiment on.\u003c/p\u003e\u003cp\u003e\u003cb\u003eThe main contributions of this study are as follows\u003c/b\u003e:\u003c/p\u003e\u003cp\u003e\u003col\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eThe various existing research on biometric authentication both unimodal and multimodal, and challenges faced on accurate authentication are highlighted\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eWe developed an Improved hybrid algorithm using machine learning and deep learning, the suggested algorithm achieved a high Accuracy.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eComprehensive experimental comparisons were performed with the existing algorithms, an Accuracy of 98.35% was achieved higher than the previous.\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003c/ol\u003e\u003c/p\u003e"},{"header":"2 Related works","content":"\u003cp\u003e\u003cb\u003eVenkata Ramana et al.\u003c/b\u003e ​[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]​ developed a multimodal system that integrates biometrics face, fingerprint, iris, palm, and ear using KLDA for feature reduction and deep learning for classification, achieving high accuracy. However, the complexity of simultaneously acquiring multiple biometric inputs could hinder real-world usability on mobile devices. \u003cb\u003eTahri \u0026amp; Beladgham\u003c/b\u003e ​[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e] presents a fuse facial and voice features at the feature level using SincNet and CNN architectures to achieve high identification rates and low error margins, enhancing multimodal security. However, Environmental factors like noise and lighting variability in mobile settings could significantly degrade performance.\u003c/p\u003e\u003cp\u003e\u003cb\u003eAmit Kumar et al.\u003c/b\u003e [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]The authors analyze fusion of face and gait biometrics using various ML models, demonstrating superior accuracy over unimodal approaches. However, While the fusion approach is sound, experiments were limited to static datasets. \u003cb\u003eAmber Hayat et al.\u003c/b\u003e [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]​ utilized​ a genetic algorithm to dynamically weight face, fingerprint, and ear biometric traits at the score fusion level, achieving 97% accuracy. However, while promising for robustness, the computational load of the genetic algorithm and its performance in noisy mobile settings are not addressed.\u003c/p\u003e\u003cp\u003e\u003cb\u003eDivan \u0026amp; Gulhane\u003c/b\u003e ​[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]presented a two-stage fingerprint matching system for secure voting, using minutiae-based algorithms. It emphasizes privacy protection by combining templates from two different fingerprints to lower FRR rates. However, the work remains conceptual without real-world testing. \u003cb\u003eRiseul Ryu et al.\u003c/b\u003e ​[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] This review synthesizes continuous multimodal authentication strategies, highlighting the dominance of score-level fusion and the need for greater attention to scalability and usability in real-world deployments. However, the review does not propose concrete methods for addressing mobile-specific limitations. \u003cb\u003eSrivastava et al.\u003c/b\u003e [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e] suggested a real-time face verification system that fuses Gabor Wavelet Transform (GWT) with Scale-Invariant Feature Transform (SIFT) and Speeded-Up Robust Features (SURF) to achieve better recognition performance and computational efficiency. The proposed method involves facial image decomposition into GWT-based sub-bands, SIFT or SURF feature extraction, and score-level fusion for verification. Benchmarked on PUT and ORL databases, the proposed GWT-SURF model achieved 98.75% accuracy in 3.4 seconds of run-time for every 100 images, outperforming conventional PCA, LDA, and basic SIFT/SURF approaches. The system was also determined to be robust against changes in lighting and image entropy, rendering it suitable for real-time biometric system. However, no comparative evaluation with modern deep learning systems and real-deployment test.\u003c/p\u003e\u003cp\u003e\u003cb\u003eChenhao Lin et al.\u003c/b\u003e ​[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] presented​ CrossBehaAuth, a deep learning-based system for cross-device keystroke authentication, leveraging CNN-GRU architecture and data augmentation to address session and device variation. However, the work is narrowly focused on keystroke dynamics and lacks integration with broader multimodal authentication frameworks. \u003cb\u003eAmmour et al.\u003c/b\u003e ​[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] devised a novel multimodal biometric system integrates Electrocardiogram (ECG) signals and fingerprint to counteract spoofing attacks. Using a transformer-based deep learning model, it demonstrates robust security enhancements. \u003cb\u003eJhansi Bharathi Madavarapu et al.\u003c/b\u003e ​[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] study investigates machine learning usage, particularly LSTM models, for behavioral biometrics like keystroke dynamics and mouse movement, achieving high accuracy. However, although effective, the study centres on desktop environments only.\u003c/p\u003e\u003cp\u003e\u003cb\u003eSwati Singh et al.\u003c/b\u003e ​[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] presented a broad overview of biometric technologies and the impact of artificial intelligence on fingerprint, facial, iris, and voice recognition methods. However, the survey lacks detailed exploration of multimodal fusion strategies or practical security system designs. \u003cb\u003eNoor Afiza et al.\u003c/b\u003e [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]​ This study presents a novel system for face detection that combines the Haar cascade technique with Convolutional Neural Networks (CNNs). Experimental results highlight the strong performance of this hybrid model, showing a notable improvement in accuracy compared to conventional machine learning methods. The combined use of Haar cascades and CNNs proves to be more effective in detecting faces than traditional approaches. However, despite the promising accuracy, the research lacks a thorough evaluation of key performance metrics such as precision and the F1 score, which are crucial for assessing the overall robustness and dependability of the system.\u003c/p\u003e\u003cp\u003e\u003cb\u003eZheng Hui Goh et al.\u003c/b\u003e [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]​ propose a multimodal authentication framework using IoM hashing and Alignment-Free Hashing (AFH) to secure templates, fusing face, fingerprint, iris, and finger-vein biometrics. The system offers strong security properties irreversibility, revocability, unlinkability. However, despite outstanding theoretical robustness, its scalability to resource-constrained mobile environments remains untested, posing challenges for real-world application. \u003cb\u003eYusuf Magaji et al.\u003c/b\u003e ​[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] proposed authentication system combines fingerprint and facial recognition to verify the identity of mobile device users. Access to the device is granted only if both authentication methods are successfully completed. However, facial recognition relies on connectivity to Microsoft servers to function. \u003cb\u003eHaq et al.\u003c/b\u003e ​[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e] evaluated various detection and recognition methods, for face recognition and face detection and compared them based on parameters like accuracy, speed and computational complexity. Strength and limitations of the methods were discussed in the review paper, with future research direction focused improving algorithm robustness, developing lightweight models for mobile devices and addressing dataset biases to enhance fairness and efficiency in face recognition. However, no clear suggestion on how to improve the alogrithms.\u003c/p\u003e"},{"header":"3 Methodology","content":"\u003cp\u003eThe proposed multimodal biometric authentication system integrates facial and fingerprint modalities to enhance identity verification accuracy and robustness, especially in scenarios where unimodal system fails due to occlusion, spoofing and environmental noise. The methodological framework consists of four key stages: data pre-processing, feature extraction via Convolution Neural Network (CNN), multimodal fusion, and classification. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates the proposed system\u0026rsquo;s flowchart\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe figure illustrates the overall process of how the proposed multimodal biometric authentication. I) sensors layer where is the use will input both this face and fingerprint. II) preprocessing layer ensures the raw data were properly normalized and ROI features are extracted. III) feature extraction layer, identifies and isolates relevant visual patterns. IV) Matching layer, used to compared the inputted biometrics with the biometrics stored in the database. V) fusion score layer, the metrics used to evaluate the quality of the image inputted. Once the fusion score reached a high level the user is granted access to the device \u0026ldquo;Authenticate\u0026rdquo;.\u003c/p\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e3.1 Pre-processing\u003c/h2\u003e\u003cp\u003ePre-processing is a modality-specific to ensure that raw data were properly normalized and that region-of-interest (ROI) features were clearly extracted. This involved several steps, such as cleaning, transforming, and normalizing the data, all aimed at enhancing the accuracy and performance of the model.\u003c/p\u003e\u003cdiv id=\"Sec5\" class=\"Section3\"\u003e\u003ch2\u003e3.1.1 Face Detection using Haar Cascade\u003c/h2\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eThe Haar Cascade method is a widely adopted approach for object detection within images, regardless of variations in scale or position. Introduced by Viola and Jones, this real-time and computationally efficient algorithm utilizes Haar feature-based cascade classifiers. The technique relies on machine learning principles, wherein a cascade function is trained using a substantial database comprising both negative and positive image samples. Haar features effectively capture grayscale variations in facial images, as they are derived from the differences in pixel intensity values, as demonstrated by Zhang et al. [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e​ Facial images were processed using the (haarcascade_frontalface_default.xml) Haar Cascade classifier from OpenCV based on the Viola-Jones algorithm. Each RGB image was first converted to grayscale for computational efficiency, and facial regions were detected using the detectMultiScale method. The first detected face (if any) was extracted as the region of interest (ROI), resized to 128\u0026times;128 pixels, and normalized to intensity values between 0 and 1. This method enables consistent face localization under varying lighting and pose conditions, providing a clean input for the CNN. This method effectively standardizes the facial input, mitigating challenges posed by background noise, varying illumination, and scale differences.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section3\"\u003e\u003ch2\u003e3.1.2 Fingerprint Minutiae Extraction\u003c/h2\u003e\u003cp\u003eA minutiae-based fingerprint recognition algorithm emphasizing local feature extraction. Initial preprocessing includes intensity normalization based on a global threshold, followed by binarization using a NOT operation to invert grayscale values, ensuring ridge pixels are represented as 1s and furrows as 0s. Morphological OPEN and CLOSE operations are applied to extract the Region of Interest (ROI) and suppress background noise, while ridge thinning reduces ridge structures to a single-pixel width without altering topology. Minutiae points are extracted using a 3\u0026times;3 window, identifying bifurcations when the center pixel with value 1 is adjacent to exactly three neighboring 1s. To mitigate false detections, spatial redundancy is reduced by eliminating bifurcation points within one-pixel proximity based on four-neighbor analysis. False minutiae are eliminated during post-processing by computing the average inter-ridge distance \u003cem\u003eD\u003c/em\u003e and comparing it with the distance \u003cem\u003ed\u003c/em\u003e between minutiae pairs. If \u003cem\u003ed\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;\u003cem\u003eD\u003c/em\u003e and both points lie on the same ridge, they are considered spurious and removed final fingerprint recognition ​[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e\u003cp\u003eFingerprint samples were preprocessed through grayscale A global thresholding operation (cv2.THRESH_BINARY_INV) was used to segment foreground ridge structures. The binarized fingerprint was then skeletonized using morphological thinning (skimage.morphology.skeletonize) to reduce ridges to a single-pixel width for accurate feature localization.Minutiae points including ridge endings and bifurcations were extracted using the Harris corner detector and corner_peaks. The detected minutiae were rendered onto a blank canvas to emphasize discriminatory structures. All images were resized to 128\u0026times;128 pixels to match the face input dimensions.\u003c/p\u003e\u003cp\u003eThis process not only improves the signal-to-noise ratio but also converts complex biometric patterns into simplified, high-information-density formats suitable for CNN-based learning.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e3.2 Feature Extraction using Convolutional Neural Networks (CNN)\u003c/h2\u003e\u003cp\u003eConvolutional Neural Networks (CNNs) have significantly transformed the fields of image analysis and computer vision by emulating the functionality of the human visual system to recognize complex visual features. These networks are composed of several key layers: convolutional layers filter-based for extracting patterns from input data, pooling layers that compress the spatial dimensions of the resulting feature maps, and activation functions that incorporate non-linear transformations. Through training methods such as backpropagation and gradient descent, CNNs are capable of automatically learning the most effective filters from the data itself, thereby eliminating the need for manual feature engineering. The versatility has enabled breakthroughs in fields such as object recognition, video analysis, and medical imaging. In this study, facial and fingerprint images were integrated into the CNN model due to its robust capabilities in handling image-based data [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]​.\u003c/p\u003e\u003cp\u003eDeep hierarchical characteristics are extracted from both modalities, using a dual-branch CNN architecture. Each branch independently processes one modality using shared architectural design principles.\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section3\"\u003e\u003ch2\u003e3.2.1 CNN Architecture\u003c/h2\u003e\u003cp\u003eEach input (face or fingerprint) passed through the following layers:\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eInput Layer\u003c/b\u003e: Accepts a tensor of shape (128, 128, 1).\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eConvolutional Layers\u003c/b\u003e: Three blocks of 2D convolution with ReLU activation and same padding were used to capture edge, texture, and region-based features. Each block contained 32, 64, and 128 filters, respectively.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eBatch Normalization\u003c/b\u003e: Applied after each convolutional layer to stabilize learning and improve convergence.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eMax Pooling\u003c/b\u003e: Reduces spatial dimensions while preserving dominant features.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eFlatten\u0026thinsp;+\u0026thinsp;Dense Layers\u003c/b\u003e: The final convolutional outputs were flattened and passed through a fully connected layer of 512-neuron with ReLU activation and dropout regularization (0.5) to avoid overfitting.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cp\u003eEach branch independently learned modality-specific representations while maintaining a consistent network depth and filter sizes for compatibility.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec9\" class=\"Section3\"\u003e\u003ch2\u003e3.2.2 Feature Fusion and Multimodal Classification\u003c/h2\u003e\u003cp\u003eThe flattened outputs of both CNN branches were concatenated to form a unified feature vector. This early fusion approach allowed the model to learn cross-modal dependencies. The fused vector passed through a shared 512-neuron fully connected layer, followed by a dropout regularization (0.5) to mitigate overfitting and softmax output layer for final classification. The Adam optimizer with categorical cross-entropy loss was used to train the system and evaluated on a hold-out test set.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003e3.3 Dataset and Implementation\u003c/h2\u003e\u003cp\u003eAll face and fingerprint images were matched by subject IDs to form multimodal pairs. Both datasets were pre-processed and split using an 80% for training and 20% for testing. TensorFlow was used to train the model for 20 epochs with 32 batch size, with automatic prefetching and batching via tf.data.Dataset. A checkpointing mechanism ensured the best-performing model on the validation set was saved. The final model was evaluated using categorical accuracy and saved in .keras format. These datasets ensure diversity in biometric characteristics and are suitable for evaluating a real-world authentication system.\u003c/p\u003e\u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eFace Dataset\u003c/b\u003e: Georgia Tech face database (gt_db), Essex face database was used, with subject-wise folders for label annotation, images are stored in a directory structure for supervised learning.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003e\u003cb\u003eFingerprint Dataset\u003c/b\u003e: FVC2000 and SOCOFing dataset was used, containing real and altered fingerprint images across multiple classes, categorized by subject and alteration types\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section3\"\u003e\u003ch2\u003e3.3.1 Essex face database [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/h2\u003e\u003cp\u003eThe public datasets utilized in this study have also been employed in prior researchers. The Faces94 dataset includes 180x200 pixel images of 153 individuals, featuring a green background, varied head orientations, minimal facial expression changes, and a range of poses. The Faces95 dataset comprises images of 72 individuals, also sized at 180x200 pixels, with frontal views and a red background created using flames. In the Faces96 dataset, there are 152 individual's images, each with a resolution of 196x196 pixels. Additionally, the Grimace dataset has 18 individual's images at 180x200 pixel resolution. To standardize input dimensions and reduce file size while preserving image quality, all images were resized to 128x128 pixels.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section3\"\u003e\u003ch2\u003e3.3.2 SOCOFing dataset​[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/h2\u003e\u003cp\u003eThe SOCOFing dataset comprises six thousand fingerprint images from six hundred African individuals, each contributing ten fingerprints. All participants are aged 18 and above. The dataset is enriched with metadata, including labels for gender, hand orientation, and specific finger identification. In addition to the original images, synthetically modified versions are included, containing z-cut, central rotation, and three different levels of modification obliteration produced with the STRANGE toolkit. The original fingerprint impressions were captured using Hamster Plus and SecuGen fingerprint scanners. Each image is formatted in grayscale with a resolution of 1 \u0026times; 96 \u0026times; 103 (gray \u0026times; width \u0026times; height).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section3\"\u003e\u003ch2\u003e3.3.3 Georgia Tech face database (Gt_Db)​ [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/h2\u003e\u003cp\u003eThe Georgia Tech face database (Gt_Db) comprises 50 classes, each containing 15 face images. These images were captured at various times, without a specific lighting setup, and feature complex backgrounds. The data collection was conducted in stages over approximately five months. The sample face images in the Gt_Db are sized at 640\u0026times;480.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section3\"\u003e\u003ch2\u003e3.3.3 FVC2000 [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]​\u003c/h2\u003e\u003cp\u003eFVC2000, the \"universal\" sensor comprises four different sensors to reflect advancements in fingerprint sensing. Small, inexpensive optical and capacitive sensors were used to collect data for Databases 1 and 2, respectively. Database 3 used a large-area optical sensor with a high quality whereas, Database 4 included artificially created images. For algorithm tuning, each database had 880 fingerprints from 110 fingers, divided into a training-set of 80 photos and a test-set of 800 images, set B and set A respectively.\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"4. Result","content":"\u003cp\u003eThis section explains the result obtained in this study, both unimodal and multimodal comparison are presented.\u003c/p\u003e\u003cp\u003eThe multimodal biometric authentication system was developed using a Convolutional Neural Network (CNN). The implementation was carried out in Python using the OpenCV open-source library. Key performance indicators like accuracy, False Acceptance Rate (FAR), and False Rejection Rate (FRR) are used to assess the system's effectiveness. Experimental testing was conducted using both unimodal and multimodal authentication approaches across various datasets.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eUnimodal Face authentication\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAuthentication Method\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAccuracy (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFRR (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFAR (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eFace Authentication (Georgia Tech face database) Gt_db\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e​​[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]98.78\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.22\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.98\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e​​[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]​94.29\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e[Our method]99.66\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.339\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.001\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe performance of unimodal face authentication presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e above, demonstrates the effectiveness of the face recognition system. The system demonstrated a high accuracy rate of 99.66%, accompanied by 0.34% of False Rejection Rate (FRR) and an exceptionally 0.001% low False Acceptance Rate (FAR). These results highlight how well the proposed face authentication method performs, surpassing earlier approaches in both accuracy and error rates. The low FAR is particularly important for ensuring that unauthorized users are not mistakenly accepted, while the low FRR indicates that legitimate users are unlikely to be rejected.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eUnimodal Fingerprint authentication\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAuthentication Method\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAccuracy (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eFRR (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFAR (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\u003cp\u003eFingerprint\u003c/p\u003e\u003cp\u003e(Sokoto fingerprint dataset) SOCOFing\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e​​[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]99.78\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e​​[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]​97.47\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e-\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e[Our method]100.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0.00\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0.00\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eThe above Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e2\u003c/span\u003e, displays the performance of unimodal fingerprint authentication using the SOCOFing dataset, which achieved 100% accuracy, with zero FRR and zero FAR. This result underscores the robustness and reliability of fingerprint recognition, making it an excellent candidate for secure authentication. The perfect accuracy and zero error rates validate the effectiveness of the minutiae-based and CNN algorithm used for fingerprint matching, ensuring both high security and user convenience.\u003c/p\u003e\u003cp\u003eDespite the high performance of both unimodal systems, challenges such as vulnerability to spoofing attacks (e.g., fake fingerprints or facial images) and environmental factors (e.g., lighting for face recognition, skin quality for fingerprint matching) persist. These drawbacks emphasize the necessity of a more thorough strategy that incorporates both modalities.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eMultimodal Biometric Authentication\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eDatabase\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eAuthentication Method\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eAccuracy (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eFRR (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eFAR (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eGt_db and TIMIT\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRef​[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e90.28\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e9.72\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e9.29\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLabelled Faces in the Wild (LFW) and FVC2000\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eRef​[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e96.26\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e3.64\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.0019\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003egt_db and FVC2000\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOur Method\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e91.12\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e8.88\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.39\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eEssex face and SOCOFing\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eOur Method\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e\u003cp\u003e98.35\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e\u003cp\u003e1.658\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e\u003cp\u003e0.01\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eWhen combining both face recognition and fingerprint verification, the multimodal biometric authentication system showed remarkable improvement in performance, as presented in Table\u0026nbsp;3 above. The multimodal system attained 98.35% accuracy, with 1.658% FRR and 0.01% FAR. These results indicate that multimodal fusion enhances the overall accuracy, ensures a very low False Acceptance Rate (FAR), and reduces False Rejection Rates (FRR). The low FAR is especially significant, as it indicates the system's high resistance to false acceptances, which is essential for securing sensitive applications such as mobile devices and access control systems.\u003c/p\u003e\u003cp\u003eThe multimodal approach benefits from the complementary strengths of face recognition and fingerprint verification. Face recognition, while non-intrusive, can be affected by poor lighting or occlusions, while fingerprint recognition may struggle with finger placement issues. By combining these modalities, the multimodal system compensates for the weaknesses of each individual modality, leading to a more robust and secure authentication solution.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn conclusion, a novel multimodal biometric authentication was introduced by combining face recognition and fingerprint verification, while for feature extraction Convolutional Neural Networks (CNNs) was used. The integration of these two biometric modalities proved effective in improving system\u0026rsquo;s reliability, security, and accuracy. Experimental findings revealed that the face-only authentication achieved a notable accuracy of 99.66%, accompanied by minimal False Rejection Rate (FRR) and False Acceptance Rate (FAR). Meanwhile, fingerprint-authentication only attained a perfect accuracy of 100%, establishing a strong performance standard for fingerprint-based authentication systems.\u003c/p\u003e\u003cp\u003eThe multimodal system further enhanced these results, achieving 98.35% accuracy, with significantly lower FRR (1.658%) and FAR (0.01%) compared to traditional unimodal methods. This improvement demonstrates the effectiveness of modality fusion, where the strengths of face recognition and fingerprint verification complement each other, compensating for their respective weaknesses in real-world conditions such as occlusion, lighting variability, and finger quality issues.\u003c/p\u003e\u003cp\u003eThe use of CNNs in feature extraction was critical to the system's success, enabling the automatic learning of optimal features from raw biometric data without requiring manual intervention. This approach not only improved the model's generalization ability but also enhanced its performance in dynamic environments.\u003c/p\u003e\u003cp\u003eDespite the strong results, challenges such as computational complexity and the quality of input data remain. Further research is needed to optimize the system for processing in real time on mobile devices and resource-constrained environments. Additionally, integrating behavioral biometrics could further strengthen the system's security and adaptability.\u003c/p\u003e\u003cp\u003eIn summary, the study's findings highlight how successful multimodal biometric systems in providing secure and dependable authentication across diverse applications, such as mobile devices. By harnessing deep learning techniques and combining multiple biometric modalities, this research provides a strong basis for future developments in biometric security technologies.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgement:\u003c/strong\u003e This study is supported by Petroleum Trust Development Fund (PTDF) Nigeria, Universiti Putra Malaysia. We are really grateful for the facilities and funding provided, which made the publication of this study possible.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions:\u0026nbsp;\u003c/strong\u003eUsman Abdul Gimba: Manuscript writing, methodology and data analysis, Noor Afiza Binti Muhd Ariffin: Manuscript reviewing, supervision, Nur Izura Udzir: Manuscript reviewing, supervision, Nor Fazlida Mohd Sani: Manuscript reviewing, supervision. All authors reviewed the manuscript and approved the final version of the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of Data and Materials:\u0026nbsp;\u003c/strong\u003eThe data used for this study can be obtained from the corresponding author upon request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Approval:\u0026nbsp;\u003c/strong\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of Interest:\u0026nbsp;\u003c/strong\u003eNo conflict of interest regarding this study shown by the authors\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical Trial Number:\u0026nbsp;\u003c/strong\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to Participate:\u0026nbsp;\u003c/strong\u003eNot applicable.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to Publish:\u0026nbsp;\u003c/strong\u003eNot applicable.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eGimba, U. A., \u0026amp; Ariffin, N. A. M. (2025). Review on User Authentication on Mobile Devices. \u003cem\u003eJournal of Advanced Research in Applied Sciences and Engineering Technology\u003c/em\u003e, \u003cem\u003e46\u003c/em\u003e(2), 26\u0026ndash;36.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOlade, I., Liang, H., ning, Fleming, C.. A Review of Multimodal Facial Biometric Authentication Methods in Mobile Devices and Their Application in Head Mounted Displays. In:, SmartWorld, I. E. E. E., Intelligence, U., Computing, Advanced, Computing, T., Computing, S., Communications, \u0026amp; Cloud (2018). \u0026amp; Big Data Computing, Internet of People and Smart City Innovation (SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI) [Internet]. IEEE; 2018. pp. 1997\u0026ndash;2004. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://ieeexplore.ieee.org/document/8560312/\u003c/span\u003e\u003cspan address=\"https://ieeexplore.ieee.org/document/8560312/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eEl-Rahiem, B. A., El-Samie, F. E. A., \u0026amp; Amin, M. (2022). Multimodal biometric authentication based on deep fusion of electrocardiogram (ECG) and finger vein. \u003cem\u003eMultimedia Systems\u003c/em\u003e (pp. 1325\u0026ndash;1337). Springer Science and Business Media Deutschland GmbH.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSengar, S., Singh, U., Hariharan, K., \u0026amp; Rajkumar (2020). International Conference on Emerging Smart Computing and Informatics (ESCI): AISSMS Institute of Information Technology, Pune, India. Mar 12\u0026ndash;14, 2020. IEEE; 2020. 309\u0026ndash;312 p.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAlshardan, A., Kumar, A., Alghamdi, M., Maashi, M., Alahmari, S., Alharbi, A. A. K. (2024). Multimodal biometric identification: leveraging convolutional neural network (CNN) architectures and fusion techniques with fingerprint and finger vein data. \u003cem\u003ePeerJ Comput Sci\u003c/em\u003e. ;10.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDargan, S., \u0026amp; Kumar, M. (2020). \u003cem\u003eA comprehensive survey on the biometric recognition systems based on physiological and behavioral modalities\u003c/em\u003e (Vol. 143). Elsevier Ltd. Expert Systems with Applications.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSajja, S. L., Mukesh, S., Hussein, A. H. A., Sunil, G., \u0026amp; Habelalmateen, M. I. (2023). Multimodal Biometric Authentication System using Probabilistic Fuzzy based Tuna Search Optimization. In: IEEE 1st International Conference on Ambient Intelligence, Knowledge Informatics and Industrial Electronics, AIKIIE 2023. Institute of Electrical and Electronics Engineers Inc.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMadduluri, S., \u0026amp; Kumar, T. K. (2024). Multimodal Biometric Authentication System using ECG Signals and Fingerprints. In: 4th International Conference on Sustainable Expert Systems, ICSES 2024 - Proceedings. Institute of Electrical and Electronics Engineers Inc.; pp. 115\u0026ndash;21.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYadav, A. K., Pateriya, R. K., Gupta, N. K., Gupta, P., Saini, D. K., Alahmadi, M., \u0026amp; Computers (2022). \u003cem\u003eMaterials and Continua\u003c/em\u003e. ;72(2):2697\u0026ndash;2712.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVenkata Ramana, S., AN ENHANCED MULTI-MODAL BIOMETRIC AUTHENTICATION SYSTEM, S. V. R., \u0026amp; USING MODIFIED DEEP LEARNING MODEL. (2023). \u003cem\u003eJournal of Science and Technology\u003c/em\u003e. ;8(12):147\u0026ndash;115.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMerit, K., \u0026amp; Beladgham, M. (2024). Enhancing Biometric Security with Bimodal Deep Learning and Feature-level Fusion of Facial and Voice Data. \u003cem\u003eJournal of Telecommunications and Information Technology\u003c/em\u003e. ;(4):31\u0026ndash;42.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKumar, A., Jain, S., \u0026amp; Kumar, M. (2024). Comparative Study of Multi-Biometrics Authentication Using Machine Learning Algorithms. In: 2024 11th International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions), ICRITO 2024. Institute of Electrical and Electronics Engineers Inc.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHayat, A., Kumar, A., Kumarbhateja, A., Pal, S. K.. An Approach for Multimodal Biometric Authentication using Genetic Algorithm. In: 15th International Conference on Computing Communication and, \u0026amp; Technologies, N. (2024). ICCCNT 2024. Institute of Electrical and Electronics Engineers Inc.; 2024.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDivan, T. A Fingerprint Matching Technique using Minutiae based Algorithm for Voting System: A Survey.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRyu, R., Yeom, S., Kim, S. H., \u0026amp; Herbert, D. (2021). \u003cem\u003eContinuous Multimodal Biometric Authentication Schemes: A Systematic Review\u003c/em\u003e (Vol. 9, pp. 34541\u0026ndash;34557). IEEE Access. Institute of Electrical and Electronics Engineers Inc.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSrivastava, R., Tomar, R., Sharma, A., Dhiman, G., Chilamkurti, N., \u0026amp; Kim, B. G. (2021). Real-time multimodal biometric authentication of human using face feature analysis. \u003cem\u003eComputers Materials and Continua\u003c/em\u003e. ;69(1).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLin, C., He, J., Shen, C., Li, Q., \u0026amp; Wang, Q. (2023). CrossBehaAuth: Cross-Scenario Behavioral Biometrics Authentication Using Keystroke Dynamics. \u003cem\u003eIEEE Trans Dependable Secure Comput\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(3), 2314\u0026ndash;2327.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAmmour, N., Bazi, Y., \u0026amp; Alajlan, N. (2023). Multimodal Approach for Enhancing Biometric Authentication. \u003cem\u003eJ Imaging\u003c/em\u003e. ;9(9).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMadavarapu, J. B., Mittal, M., Salagrama, S., Adnan, M. M., Rana, A., \u0026amp; Yadav, K. (2024). Behavioral Biometrics Authentication Systems: Leveraging Machine Learning for Enhanced Cybersecurity. In: Proceedings of International Conference on Communication, Computer Sciences and Engineering, IC3SE 2024. Institute of Electrical and Electronics Engineers Inc.; pp. 1478\u0026ndash;83.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSingh, S., Sharma, S., Awasthi, M., Rawat, S., \u0026amp; Chanti, Y. (2024). Advancements of Emerging Technologies in Biometrics Authentication. In: 2024 IEEE 1st Karachi Section Humanitarian Technology Conference, Khi-HTC 2024. Institute of Electrical and Electronics Engineers Inc.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMohd Ariffin, N. A., Gimba, U. A., \u0026amp; Musa, A. (2025). Face Detection based on Haar Cascade and Convolution Neural Network (CNN). Journal of Advanced Research in Computing and Applications [Internet]. ;38(1):1\u0026ndash;11. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.akademiabaru.com/submit/index.php/arca/article/view/5576\u003c/span\u003e\u003cspan address=\"https://www.akademiabaru.com/submit/index.php/arca/article/view/5576\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGoh, Z. H., Wang, Y., Leng, L., Liang, S. N., Jin, Z., Lai, Y. L., et al. (2022). A Framework for Multimodal Biometric Authentication Systems With Template Protection. \u003cem\u003eIeee Access : Practical Innovations, Open Solutions\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e, 96388\u0026ndash;96402.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYusuf, M., Abdul Gimba, U., Uthman Bello, A., Habu Adamu, A., \u0026amp; Salisu, S. (2019). Two Way Authentication for Android Mobile Phones. Vol. 5, Dutse Journal of Pure and Applied Sciences (DUJOPAS). Jun.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eUl Haq, M., Sethi, M. A. J., Ahmad, S., Ahmad, N., Anwar, M. S., \u0026amp; Kutlimuratov, A. (2025). \u003cem\u003eA Comprehensive Review of Face Detection/Recognition Algorithms and Competitive Datasets to Optimize Machine Vision\u003c/em\u003e (Vol. 84, pp. 1\u0026ndash;24). Tech Science. Computers, Materials and Continua.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang, X., Cheng, D., Jia, P., Dai, Y., \u0026amp; Xu, X. (2020). An Efficient Android-Based Multimodal Biometric Authentication System with Face and Voice. \u003cem\u003eIeee Access : Practical Innovations, Open Solutions\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e, 102757\u0026ndash;102772.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNarwal, S., \u0026amp; Kaur, D. (2016). Comparison between Minutiae Based and Pattern Based Algorithm of Fingerprint Image. International Journal of Information Engineering and Electronic Business [Internet]. ;8(2):23\u0026ndash;9. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.mecs-press.org/ijieeb/ijieeb-v8-n2/v8n2-3.html\u003c/span\u003e\u003cspan address=\"http://www.mecs-press.org/ijieeb/ijieeb-v8-n2/v8n2-3.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSalturk, S., \u0026amp; Kahraman, N. (2024). Deep learning-powered multimodal biometric authentication: integrating dynamic signatures and facial data for enhanced online security. \u003cem\u003eNeural Computing And Applications\u003c/em\u003e, \u003cem\u003e36\u003c/em\u003e(19), 11311\u0026ndash;11322.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAminu, M., \u0026amp; Ahmad, N. A. (2022). Locality preserving partial least squares discriminant analysis for face recognition. \u003cem\u003eJournal of King Saud University - Computer and Information Sciences\u003c/em\u003e, \u003cem\u003e34\u003c/em\u003e(2), 153\u0026ndash;164.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIsah Shehu, Y., Ruiz-Garcia, A., Palade, V., \u0026amp; James, A. Sokoto Coventry Fingerprint Dataset [Internet]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.kaggle.com/\u003c/span\u003e\u003cspan address=\"https://www.kaggle.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003e\u0026ccedil;oskun, M., U\u0026ccedil;ar, A., Yildirim, \u0026Ouml;., Demir, Y.. MEES 2017: proceedings of the International Conference on Modern Electrical and Energy SystemsKremenchuk Mykhailo Ostrohradskyi National University, Ukraine, 15\u0026ndash;17, \u0026amp; November (2017). IEEE; 2018. 376\u0026ndash;379 p.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMaio, D., Maltoni, D., Cappelli, R., Wayman, J. L., \u0026amp; Jain, A. K. (2000). FVC: Fingerprint Verification Competition.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHamou, A. O., \u0026amp; Chelali, F. Z. (2024). Facial Recognition Based on Machine and Deep Learning. In: ICAEE 2024 - International Conference on Advanced Electrical Engineering 2024. Institute of Electrical and Electronics Engineers Inc.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSingh, R., Sharma, N., Chauhan, R., Choudhary, A., \u0026amp; Gupta, R. (2023). Enhanced Fingerprint Alteration Detection Using Lightweight CNN Model Trained on SOCOFing Dataset. In: 2023 3rd International Conference on Smart Generation Computing, Communication and Networking, SMART GENCON 2023. Institute of Electrical and Electronics Engineers Inc.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNarayanan, A., \u0026amp; Hameeduddin, Q. M. (2023). Gender Detection and Classification from Fingerprints Using Convolutional Neural Network. In: ICSPC 2023\u0026ndash;4th International Conference on Signal Processing and Communication. Institute of Electrical and Electronics Engineers Inc.; pp. 178\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJena, P. P., Kattigenahally, K. N., Nikitha, S., Sarda, S., \u0026amp; Harshalatha, Y. (2021). Multimodal Biometric Authentication: Deep Learning Approach. In: 2021 International Conference on Circuits, Controls and Communications, CCUBE 2021. Institute of Electrical and Electronics Engineers Inc.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-computing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Discover Computing](https://link.springer.com/journal/10791)","snPcode":"10791","submissionUrl":"https://submission.springernature.com/new-submission/10791/3","title":"Discover Computing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Unimodal authentication, Multimodal authentication, CNN, Face recognition, Fingerprint","lastPublishedDoi":"10.21203/rs.3.rs-7048774/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7048774/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eA novel multimodal biometric authentication system combining face and fingerprint verification to ensure enhanced security, accuracy, and resilience in user identification, is presented in this work. The system utilizes Convolutional Neural Networks (CNNs) for effective feature extraction from both biometric modalities, addressing challenges such as occlusion, lighting, and finger quality in real-world scenarios. The results prove the performance of the system, with unimodal face authentication achieving 99.66% accuracy, unimodal fingerprint authentication reaching 100% accuracy, and the multimodal system is 98.35% accurate overall. The multimodal approach significantly reduces False Rejection Rate (FRR) and False Acceptance Rate (FAR), enhancing authentication security and user convenience. The combination of fingerprint and face modalities allows for improved performance by compensating for the weaknesses of individual modalities. The study highlights the potential of deep learning in biometric systems, providing a robust solution for secure access control in mobile and high-security applications. Future work will focus on the further enhancement of the system, for real-time deployment on mobile devices, expanding the model's applicability across diverse environments, and exploring the integration of additional biometric modalities.\u003c/p\u003e","manuscriptTitle":"Enhancing Biometric Authentication through Multimodal Approach Combining Face and Fingerprint Recognition Using Convolutional Neural Networks (CNN)","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-22 10:45:25","doi":"10.21203/rs.3.rs-7048774/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-09-02T05:56:48+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-29T23:23:16+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-08-20T06:58:34+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"107460267869932911622145920082116366907","date":"2025-08-17T13:53:06+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"151901439011944818993924325086657961357","date":"2025-08-14T10:03:18+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-08-14T09:18:44+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-08-14T09:00:35+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-08-07T16:56:02+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-08-02T18:12:35+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Computing","date":"2025-08-02T17:57:30+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-computing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Discover Computing](https://link.springer.com/journal/10791)","snPcode":"10791","submissionUrl":"https://submission.springernature.com/new-submission/10791/3","title":"Discover Computing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"369648fa-c006-46b1-80eb-f500ea3c9221","owner":[],"postedDate":"August 22nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-11-03T16:06:26+00:00","versionOfRecord":{"articleIdentity":"rs-7048774","link":"https://doi.org/10.1007/s10791-025-09775-z","journal":{"identity":"discover-computing","isVorOnly":false,"title":"Discover Computing"},"publishedOn":"2025-10-30 15:58:54","publishedOnDateReadable":"October 30th, 2025"},"versionCreatedAt":"2025-08-22 10:45:25","video":"","vorDoi":"10.1007/s10791-025-09775-z","vorDoiUrl":"https://doi.org/10.1007/s10791-025-09775-z","workflowStages":[]},"version":"v1","identity":"rs-7048774","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7048774","identity":"rs-7048774","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}