A Novel Approach to the Extraction of Iris Features Using a Deep Learning Network

A Novel Approach to the Extraction of Iris Features Using a Deep Learning Network | 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 A Novel Approach to the Extraction of Iris Features Using a Deep Learning Network Poovayar Priya M, Ezhilarasan M This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5345891/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 6 You are reading this latest preprint version Abstract Iris recognition is a powerful biometric identification technique due to the uniqueness and complexity of iris patterns. However, existing methods often struggle to capture both global and local iris features, which are essential for accurate and reliable recognition. Specifically, details like lacunae, Wolfflin nodules, contraction furrows, and pigment spots hold significant information about an individual's identity, yet accurately extracting and integrating these features remains challenging. This study introduces the Iris-FE Architecture, a convolutional neural network (CNN)-based solution designed to address these challenges by extracting both global and local iris features in a unified framework. The purpose of this study is to develop and evaluate a CNN architecture, Iris-FE, that captures and combines global and local iris features for more accurate and comprehensive iris recognition. By leveraging advanced feature extraction techniques, this research aims to improve the reliability and performance of iris recognition systems and enhance their practical applications. The proposed Iris-FE Architecture uses Global features (lacunae and Wolfflin nodules) that are fed into the Iris Surface Global Features (ISGF) model, while local features (contraction furrows and pigment spots) are processed through the Iris Surface Local Features (ISLF) model. The outputs of ISGF and ISLF are combined to form a comprehensive feature set, enhancing the system's ability to capture hierarchical patterns and fine-grained details. The Iris-FE Architecture successfully demonstrated its capability to accurately extract and integrate both global and local iris features. Testing on the iris image dataset showed a notable improvement in identification accuracy and robustness compared to traditional feature extraction methods. The model consistently outperformed baseline approaches, particularly in complex images where both global textures and intricate local features were crucial for accurate recognition. Iris Recognition Feature extraction Iris surface features CNN Deep learning Iris Surface Global Features Iris Surface Local Features Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 INTRODUCTION The iris is a sensitive, ring-like structure located between the sclera and the pupil. It plays a crucial role in controlling the size of the pupil, thus managing the amount of light that enters the retina. The iris, which encircles the pupil, exhibits unique features such as furrows, pits, and crypts that are specific to each person and remain consistent throughout life, even in identical twins. Unlike fingerprints, the iris is more resistant to forgery due to the protective sclera. Iris recognition is a sophisticated biometric technique that utilizes the visible iris for authentication. This allows for non-invasive capture of images from distances of 7.5 to 15 cm (3 to 10 inches). The iris’s enduring and intricate patterns make it a highly reliable biometric identifier [1]. Since the iris image obtained from the sensor is typically raw and unprocessed, it is not directly used for authentication. Instead, the iris's inherent textures and patterns known for their uniqueness are utilized for identification purposes [2]. The iris is a complex structure with distinctive features such as the pupil, pupillary zone, collarette, ciliary zone, radial furrows, and crypts as shown in Fig. 1 . These features indicate the intricate pattern of the iris, which and its color variations play a crucial role in identification. The process relies heavily on analyzing the intricate textures of the iris, which contain rich spatial frequency information. Techniques like Gabor filters and wavelet transforms are often used to emphasize the high-frequency details, such as crypts and furrows, that distinguish one iris from another. Multi-resolution feature extraction methods allow for the capture of both fine and coarse iris structures, ensuring that no critical detail is overlooked [3]. This ability to capture complex patterns, combined with the iris's natural protection by the sclera and eyelids, makes iris recognition highly resistant to forgery. Moreover, despite the high-dimensional nature of the data, the feature extraction process efficiently condenses the information into a compact, distinctive representation, enabling rapid and accurate matching against stored templates. This makes iris recognition one of the most secure and precise biometric identification systems [4]. The use of deep learning algorithms in iris recognition has significantly improved the accuracy and robustness of feature extraction. Traditional feature extraction methods, such as Gabor filters and wavelet transforms, rely on handcrafted features that capture specific textures of the iris. While effective, these methods may not always account for the wide variability in iris images due to noise, lighting, or occlusions. Deep learning, on the other hand, enables automatic learning of feature representations directly from the raw image data, allowing the model to capture more complex and subtle patterns in the iris that traditional methods might miss [5]. Convolutional Neural Networks (CNNs), in particular, have revolutionized iris recognition by automatically learning hierarchical features from the input image. Instead of relying on pre-defined filters, CNNs can learn optimal filters that extract the most relevant features from the iris at multiple levels of abstraction. This results in a more comprehensive and accurate representation of the iris, improving recognition performance even under challenging conditions, such as poor lighting or partial occlusion by eyelids or eyelashes [6]. Furthermore, deep learning models can generalize better to variations in iris images, such as those caused by changes in pupil size, head tilt, or noise in the image. Advanced architectures like Res-Net, VGG-Net, and MobileNetV2, when applied to iris recognition, can capture fine-grained details while being robust to distortions. These networks can also be fine-tuned to enhance specific aspects of iris feature extraction, leading to state-of-the-art accuracy and reduced error rates. Incorporating deep learning algorithms has not only improved the precision of iris recognition systems but also made them more adaptive and scalable. This ability to learn and refine features without manual intervention offers significant advantages over traditional methods, positioning deep learning as a key technology for the future of biometric recognition [7]. LITERATURE SURVEY In iris recognition, existing feature extraction techniques are generally categorized into three main types: texture-based, geometric-based, and macro-based feature extraction. Each of these approaches targets different aspects of the iris, allowing the system to capture a wide range of distinguishing details for accurate identification. 1. Texture features The iris region is divided into the pupillary zone and ciliary zone for feature extraction and encoding. The Daugman rubber sheet model is employed to map the iris texture features from these zones into a feature vector. This model transforms the entire iris region into a rectangular shape by considering the density of the pupillary zone, which is closer to the pupil and less affected by obstructions such as eyelids or lashes. The pupillary zone, with its lower frequency and richer information, is preferred for identification over the ciliary zone. Before feature extraction, the segmented iris image undergoes normalization to address iris size and lighting variations. This process is essential to counteract the effects of elastic deformation, which can impact matching performance. For iris image classification, texture features [8] are combined with Speeded Up Robust Features (SURF) [9] key points. The Descriptor Number On Limits (DENOL) method generates a fixed number of SURF key points for each image in the dataset by adjusting the Hessian threshold parameter. SURF is known for its computational efficiency and robustness to scale changes, rotation, and illumination, making it suitable for real-time applications and large-scale data processing. SURF operates through a series of Gaussian convolutions to create a scale space for multi-scale feature detection. It identifies interest points by analyzing the Hessian matrix determinants and detects local maxima to ensure stable & repeatable features. Each point of interest is analyzed within a square region, subdivided into smaller areas for descriptor extraction. Haar wavelet responses are computed in both horizontal and vertical directions for each sub-region, creating descriptor vectors that represent the spatial distribution of local image gradients. Comparing these descriptors across images involves measuring the Euclidean distance between vectors to identify matching local structures. Iris feature extraction faces performance issues in certain conditions, particularly with highly textured iris images. The structure of the texture features is as follows: Pupillary Zone Ciliary Zone Outer Edge The following is the representation of the feature vector stored in the Database: a. Size: 2048 bits (the total length of the feature vector). b. Position: Specific positions within the vector might encode different types of texture information. i. Positions 0–511: Encodes texture patterns from the pupillary zone. ii. Positions 512–1535: Encodes texture patterns from the ciliary zone. iii. Positions 1536–2047: Encodes texture patterns from the outer edge. 2. Geometric features Geometric traits of the iris are closely linked to its shape which are valuable forbiometric identification. These geometric features can be leveraged at various levelsof identification. While individual features like the pupil's size might not be highlydistinctive on their own, they become uniquely identifying when combined with othercharacteristics such as the iris's texture or the collarette. The integration of these geometricelements enhances the overall uniqueness of the iris which makes it a robustfeature for identification purposes. i. Pupil Based Geometric features Although the pupil is generally defined as circular, it is not always perfectly circular. However, the geometric characteristics still need to be recognized and identify the pupil. The following sections will consider and describe the pupil roundness, size, and smoothness. a. Pupil Roundness Several orientation points on the iris can be used to assess the consistency of its diameter, as roundness measurement involves tracing the iris 360o around. Generally, roundness is defined by the degree of the edges and corners of a circle related to the circle’s compactness and sphericity. b. Pupil Largeness The pupil radius is used to characterize its size. However, the general radius is inadequate because the pupil is not a perfect circle. Eight directions on the pupil: 0 o , 45 o , 90 o , 135 o , 180 o , 225 o , 270 o and 315 o are used to compute the radius. c. Pupil Smoothness The curvature can be used to represent the smoothness of the iris by a circle, which is defined as the reciprocal of the radius (1/r), Since the pupil is regarded as a circle, the Pupil Smoothness (PS) is measured using the curvature of the circle. ii. Collarette-Based Geometric features The collarette region, located in the inner part of the iris remains unaffected by externalfactors such as eyelids or lashes and is not influenced by pupil dilation. This areahas a fixed radius range and is concentric with the pupil. Characterized by distinct dotsor radial spokes, the collarette's geometric features must first be identified within the irisbefore any geometric measurements can be calculated. Its stable location and uniquepattern make it a reliable feature for biometric identification. iii. Geometric Features of Iris In iris identification, geometric features are crucial as they provide distinct and measurable attributes that enhance the accuracy and reliability of biometric identification systems. Key geometric parameters such as iris proximity and diameter offer valuable insights into the intricate structural details of the iris. The iris diameter, a fundamental metric in the normalization process measures the horizontal distance from the pupil's centre to the iris's edge which ensures consistency across different images. Additionally, iris proximity measures the spatial relationships between the iris, pupil and sclera which captures the radial distance from the pupil centre to the iris edge and other defining features. These geometric measurements contribute significantly to distinguishing and verifying individual irises. a. Iris diameter The average diameter of the iris at eight different locations is used to compute this characteristic. Simply measuring the distance from the iris’s center to any point on its outer edge is insufficient, as the iris is not a perfect circle. Its shape can vary, making it important to consider multiple measurements to accurately represent its dimensions. Therefore, the average iris diameter is determined on the pupil at twelve different angles: 0 o , 15 o , 30 o , 45 o , 60 o , 75 o , 90 o , 105 o , 120 o , 135 o , 150 o and 165 o . b. Iris proximity A line that crosses a circle exactly once is called a tangent to it. The term "point tangency" refers to the junction point. At the precise point of tangency, which is a straight angle of 90 o and perpendicular to the line drawn from the centre, the tangent hits the circle. From the outer point (sclera) to the iris, a tangent is drawn. The iris image was converted into a rectangular format using integral differential operators and rubber sheet methods, followed by cropping the rectangular region based on the iris map. Feature extraction was carried out using various techniques, including wavelet transform, first-order statistical analysis, Gray-Level Co-Occurrence Matrix (GLCM) [10], and Gray Level Run Length Matrix (GLRLM) [11]. Wavelet transform is particularly effective for processing signals and images, enabling multi-resolution analysis that decomposes the image into different frequency components. This approach captures both spatial and frequency information simultaneously which is essential for highlighting geometric features and ensuring spatial localization thus simplifying the detection of localized features critical for distinguishing between different irises. First-order statistical analysis computes basic image statistics such as average, standard deviation, median, and minimum & maximum values. These statistics describe the image’s intensity distribution without considering spatial relationships, offering insights into the overall geometric features and intensity variations of the iris. The GLCM is a statistical method used to examine spatial relationships between pixels in an image. It captures geometric features through metrics such as contrast, correlation, energy, homogeneity, and entropy by analyzing how frequently specific pairs of pixel intensities occur at various distances and angles. GLCM is effective in detecting repetitive directional patterns in the iris which are useful for texture analysis in iris identification. The GLRLM estimates the size of continuous pixels with similar gray levels at given angles. Attributes from GLRLM such as short-run emphasis, long-run emphasis, gray level non-uniformity, run length non-uniformity and run percentage help to determine the roughness and orientation of geometric features. GLRLM is valuable for characterizing the iris as it captures directional patterns involved in geometric features across multiple directions. These systems often struggle to handle high-dimensional feature spaces, which can result in lower accuracy and efficiency. The structure of the geometric features are as follows: PR PL PS CR ID IP a. Pupil Roundness (PR) A single floating-point value representing the roundness of the pupil consists of 32 bits. b. Pupil Largeness (PL) An array of 8 floating-point values with 256 bits representing the radius at different angles (0°, 45°, 90°, 135°, 180°, 225°, 270°, 315°). c. Pupil Smoothness (PS) An array of 8 floating-point values with 256 bits representing the smoothness at Different angles which is computed as the curvature (1/radius). d. Collarette Radius (CR) An array of 12 floating-point values with 384 bits representing the radius of the Collarette at different angles (0°, 15°, 30°, ..., 165°). e. Iris Diameter (ID) An array of 12 floating-point values with 384 bits representing the average diameter of the iris at different angles (0°, 15°, 30°, ..., 165°). f. Iris Proximity (IP) An array of 8 floating-point values with 256 bits representing the proximity between the iris and the sclera at different angles. 3. Macro features The visible features that constitute the structural components of the iris and render it unique for identification are referred to as macro-level features. These macro characteristics include noticeable discontinuities in the iris’s size, color, texture, and intensity. The iris comprises numerous intricate and distinctive patterns, which can be leveraged for identification. In this context, radial furrows and crypts are particularly significant for distinguishing individuals as detailed in the following sections: i. Crypts The crypts are minute yet significant details of the iris, appear as pit-like depressions or extra apertures found around the outer edge of the ciliary region and near the collarette. These crypts facilitate the flow of aqueous humor from the stroma and deeper iris tissues. Typically, crypts are located closer to the collarette and display a darker hue compared to the surrounding iris areas. The structure of the crypts: Size Position Intensity a. Size: A floating-point value representing the average size of the crypts with 32 bits. b. Position: An array of 8 floating-point values with 256 bits representing the positions of the crypts relative to the collarette in polar coordinates (r, θ). c. Intensity: A floating-point value representing the average intensity of the crypts compared to the surrounding iris tissue with 32 bits. ii. Radial Furrows Radial furrows extend outward in all directions from the centre of the pupil, potentially starting near the pupil and extending through the collarette. These fissures in the anterior layer allow for simulation of the iris with loose tissue sometimes protruding outward. This feature is height-dependent due to the outward bulging. Despite their variable appearance, the relative positions of radial furrows remain stable. The structure of the radial furrows: Length Position Intensity a. Length: An array of 8 floating-point values with 256 bits representing the length of radial furrows measured from the pupil outward. b. Position: An array of 8 floating-point values with 256 bits representing the relative position of radial furrows along the circumference of the iris. c. Intensity: A floating-point value with 32 bits representing the intensity variation within the radial furrows compared to the surrounding iris tissue. The use of the Scale-Invariant Feature Transform (SIFT) [12] Scale-Invariant Feature Transform algorithm for detecting key points and performing iris matching is highlighted in [13]. SIFT is particularly advantageous for iris identification due to its robustness in detecting distinctive features that remain invariant to scale, rotation and illumination changes. This characteristic makes SIFT well-suited for extracting and identifying fine patterns in the iris such as crypts and radial furrows. SIFT operates by constructing a scale space using the Difference of Gaussians (DoG) approach to detect potential key points. It then localises these key points accurately and assigns orientations based on the local gradient direction which ensures rotation invariance. The final step involves generating descriptors by computing gradient magnitudes and orientations within a local region around each key point, capturing unique texture information. In iris identification, SIFT can effectively identify key points like crypts are small, irregular dark spaces resulting from iris tissue separations. Radial furrows are lines extending from the pupil. The scale and rotation invariance of SIFT combined with its robustness to illumination changes, enhances the accuracy and reliability of iris identification systems. By providing consistent and detailed feature information regardless of variations in eye orientation, image resolution, or lighting conditions. Moreover, many iris feature extraction algorithms focus on the local neighborhood of each pixel and fail to capture global spatial relationships within the iris image. MATERIALS AND METHODS Datasets: The Puducherry Technological University (PTU)-IRIS dataset is used in experiments with the proposed work. Two sessions were used to gather the iris images utilizing a Mobile-EyesTM sensor, a product of the L-1 Identify solution. There are 5000 participants in the dataset; 4000 were used for training and 1000 for testing from faculty members and students of the University. The individuals were asked directly if they wore contact lenses during the process of taking their images and this was noted. The images were taken with near-infrared (NIR) light. It records SVGA images with a width of 752 x 480 at a rate of 30 frames per second. Due to the dual-iris enrollment feature of the gadget, simultaneous auto-capture of the left and right eyes is possible. It can function in both dimly lit and brilliantly lit areas. The scanned image has a resolution of 640 x 480 and is saved in the PNG file format. Proposed Work: Feature extraction is a crucial part of an iris identification system. In the existing systems, there are three iris features such as textural features for iris patterns, geometric features for iris shape, pupil & collarette, and macro features for visual information. However, the existing features suffer from performance issues in certain conditions, particularly with highly textured iris images. These features often struggle to handle high-dimensional feature spaces which can result in lower accuracy and efficiency. Additionally, many iris feature extraction algorithms focus on the local neighborhood of each pixel and fail to capture global spatial relationships within the iris image. The iris surface features proposed in this paper address all these issues in the existing systems. In iris identification, individual differentiation depends on a thorough examination of iris surface features. These surface features include lacunae, Wolfflin nodules, Contraction furrows, and pigment spots. These distinctive iris textures and patterns are important for biometric identification. These features are compared with the existing features for better identification results. The characteristics of iris surface features are discussed in the following subsections. i. Lacunae Lacunae are distinct clustered lesions found in the iris and play a significant role in iris identification systems. During the identification process, these lacunae are detected and modeled as clustered shapes with parameters like size, orientation, and count, which are extracted as unique features of the individual's iris. Due to their unique patterns and distribution, lacunae as shown in Figure 2 contribute to creating a highly individualized biometric template. If a closed lesion is found during iris diagnosis, it is strongly advised that the patient be instructed to have a comprehensive checkup. The count of each Lacunae measure is extracted as features [14]. Figure 2: Lacunae signs in the iris ii. Wolfflin Nodules In the iris identification system, Wolfflin nodules play a critical role by providing unique biometric features for identification. During the identification process, Wolfflin nodules as shown in Figure 3 characterized by their open-ended, circular, or near-circular raised areas which are detected in the iris image. These nodules are indicative of ongoing metabolic processes and can reveal information about the state of associated organs. The count and characteristics of each Wolfflin nodule are extracted as distinct features of the iris. These features are then incorporated into the biometric template which represents the unique characteristics of an individual's iris. During identification, the extracted features from a newly captured iris image are compared against the stored templates in the database [15]. Figure 3: Wolfflin nodules signs in the iris iii. Contraction Furrows Contraction rings show how the neurological and muscular systems are related. Irritative stress causes long-lasting contractive spasms that obstruct the regular flow of fluids that bring waste products and nutrients to the muscles. The ring itself forms a valley or mountain ridge and is frequently referred to as a contraction furrow as shown in Figure 4. There are occasions when the individual iris fibers exhibit a zigzag pattern in addition to this ridge. The ridge appears to be a valley as the nerve ring becomes deeper. One or more segments of curving furrows that follow the circumference of the iris and range in color from bright white to dark reveal nerve rings also known as neurovascular cramp rings. They display a level of stress and anxiety. Nerve rings indicate tension and a desire for rest. This may come from physical or mental areas. Contraction furrows are modeled as concentric circles with varying radii. The count of each furrow is taken as a feature of Contraction furrows [16]. Figure 4: Contraction furrow signs in the iris iv. Pigment Spots Psora, often known as drug spots, are irregular pigment marks that show the buildup of drugs or products of malfunctioning metabolism as shown in Figure 5. When inherited, these characteristics are regarded as miasms or inherited stains from ancestor illnesses. Pigment spots are modeled as small circular regions with intensity or color variations. The count of each spot is considered a feature of pigment spots [17]. Figure 5: Pigment Spots signs in the iris Iris-FE Architecture for Feature Extraction The Iris-FE Architecture for feature extraction as shown in Figure 6 extracts these features and combines the global details with the local details. Convolutional neural network (CNN) architecture, Iris-FE is renowned for its efficiency and simplicity. It comprises several small (3x3) receptive field convolutional layers, followed by fully connected layers. A feature descriptor called Iris Surface Local Features (ISLF) measures how gradient orientations are distributed in specific areas of an image. The Iris-FE extracts high-level information and local details separately, combining both, it extracts both global details as well as local details. The proposed algorithm takes a normalized and enhanced grayscale iris image (Igray) as input. After that, the image is scaled to 224x224 pixels in resolution to make it compatible with the Iris-FE model which is a standard input size. In order to comply with the limitations of the model and preserve the spatial integrity of the iris features, this scaling is essential. The image is resized and then goes through a preprocessing stage where its dimensions are increased and further normalized to satisfy the deep learning model's input specifications. The feature extraction process is divided into four key features, each targeting specific characteristics of the iris. The contours are used to delineate the boundaries of various iris surface features such as lacunae, Wolfflin nodules, contraction furrows, and pigment spots. These contours are typically found by applying edge detection algorithms to a binary image which simplifies the process by reducing the image to its most basic structure. Lacunae features are extracted by detecting contours within the binary version of the iris image (Ibinary). The algorithm identifies and counts closed contours as Lacunae. Wolfflin nodules are quantified by counting the total number of contours detected providing a measure of these raised areas on the iris surface. Contraction furrows, which are large and have deep curvature are identified by detecting contours with an area exceeding a predefined threshold (‘T’). Finally, pigment spots are extracted using a Blob detection technique, where the algorithm counts the number of distinct blobs or key points detected in the binary image. Figure 6: Iris-FE Architecture for Feature Extraction Once these features are extracted, the Lacunae and Wolfflin nodules features are fed into the global feature model to extract deep features (F ISGF ). ISGF is particularly effective in capturing hierarchical and spatially complex patterns making it suitable for analyzing the intricate textures of the iris. Simultaneously, the features of the contraction furrows and pigment spots are processed through the local feature model to extract local features (F ISLF ). This model excels at capturing edge information and local textures, which are vital for detailed iris analysis. The global and local features obtained from these models are combined as the final output providing a comprehensive feature set for further analysis or classification tasks in iris identification systems. The extraction of features using Iris-FE is as follows: Algorithm: Iris-FE Step 1: Input the normalized iris image. Step 2: Resize the input image to 224x224 size resolution. I resized = resize (I gray , (224,224)) Step 3: Expand dimensions and preprocess the image for the Iris-FE. I preprocessed = preprocessInput (expandDims(I resized ,0)) Step 4: (i) Using contours, the Lacunae features are extracted from the iris Search the contours in the binary image: C = findContours(Ibinary) Count the number of closed contours: L = Σ δ (isContourConvex(c)) c ∈ C where L is the Lacunae count, and δ is an indicator function that returns 1 if the contour c is closed, and 0 otherwise. (ii) Extract the Wolfflin nodules from the iris using contours. Count all the contours in the binary image. W = |C| where |C| is the total number of contours. (iii) Extract the Contraction furrows from the iris using large contour detection. It is defined as: F = Σ c ∈ C δ (area(c) > T) where T is the predefined threshold. (iv) Extract the pigment spots from the iris using a Blob detector. Apply Blob detection to the binary image: K = blobDetector(I binary ) Count the number of key points detected: P = |K| where |K| is the number of detected key points. Step 5: Pass all the deep features to the Iris Surface Global Features Iris Surface Global Features (ISGF) model as: F ISGF = ISGF(L,W) Step 6: Pass all the local features to the Iris Surface Local Features (ISLF) model as: F ISLF = ISLF(F,P) The ‘resize’ function allows a resolution of 224x224 pixels. The reason for resizing is to ensure that the image matches the input size requirement for the Iris-FE model. These models typically expect a fixed input size, so resizing standardizes the input data. The expandDims’ function represents the model's expected inputs in the form of batch_size, height, width, and channels. The structure of different feature vectors is stored in the database as follows: i) The structure of the Lacunae Feature Vector (LFV) is given as follows: Size Position Density Distribution a. Size: Represents the size of each lacuna (e.g., area or diameter). b. Position: The (x, y) coordinates of the lacuna relative to a reference point in the normalized iris image. c. Shape: Describes the shape of the lacuna, which could be "round", "oval", "or irregular" etc. d. Distribution: Describes how the lacunae are distributed within the iris, such as "clustered", "isolated" or "dispersed". The total size required for each LFV is: Total size = Size + Position + Density + Distribution= 4 bytes + 8 bytes + 4 bytes + 1 byte=17 bytes. ii) The structure of the Wolfflin nodules Feature Vector (WFV) is given as follows: Size Position Shape a. Size: Represents the size of each nodule (e.g., area or diameter). b. Position: The (x, y) coordinates of the nodule relative to a reference point in the normalized iris image. c. Shape: Describes the shape of the nodule, which could be "round", "oval" "irregular" etc. The total size required for each WFV is: Total size = Size + Position + Shape = 4 bytes + 8 bytes + 1 byte=13 bytes iii) The structure of the Contraction Furrows Feature Vector (CFV) is given as follows: Curvature Depth Length Position a. Curvature: This represents the curvature of each furrow, which may vary depending on its location within the iris. b. Depth: Indicates the depth of each furrow c. Length: Measures the length of the furrow, which can be important for distinguishing between different furrows. d. Position: The (x, y) coordinates of the furrow relative to a reference point in the normalized iris image. The total size required for each CFV is: Total size=Curvature + Depth + Length+ Position=4 bytes+4 bytes +4 bytes+8 bytes=20 bytes. iv) The structure of the Pigment Spots Feature Vector (PFV) is given as follows: Size Position Shape a. Size: Represents the size of each spot (e.g., area or diameter). b. Position: The (x, y) coordinates of the spot relative to a reference point in the normalized iris image. c. Shape: Describes the shape of the spot, which could be "round", "oval", "irregular" etc. The total size required for each PFV is: Total size = Size + Position + Shape = 4 bytes + 8 bytes + 1 byte=13 bytes EXPERIMENTAL RESULTS AND DISCUSSION Figure 7 shows the accuracy of the acceptance rate of proposed features for the PTU-IRIS dataset. The computation of the distancesbetween biometric pairs was compared to evaluate the acceptance rates at various levels. The ROC curve also shows the trade-offs betweensecurity (low FAR) and usability (high GAR) for the MMU1-IRIS dataset. From the ROCcurve, it is inferred that the GAR values of Lacunae, Wolfflin nodules, Contraction furrows, andpigment spots are 97.5%, 86.5%, 79.8%, and 79.4% respectively. The visualization of this ROCcurve shows that the Lacunae feature outperformed others. Figure 7: ROC curve of proposed features for the PTU-IRIS dataset Figure 8 examines the DET curves for four distinct features of the PTU-IRIS dataset: Lacunae, Wolfflin nodules, Contraction furrows, and Pigment spots. The Error rates for Lacunae, Wolfflin nodules, Contraction furrows, and Pigment spots are 0.025%, 0.05 %, 0.08%, and 0.1% respectively. From this, Lacunae has a lower error rate than other features. Figure 8: DET curve of proposed features for the PTU-IRIS dataset Table 1 systematically evaluates the performance of various biometric authentication methods using the PTU-IRIS dataset focusing on the GAR as a primary metric. The dataset was examined using four distinct biometric features: Lacunae, Wolfflin Nodules, Contraction Furrows, and Pigment Spots. The GAR, which measures the percentage of genuine pairs correctly accepted by the system, varied across the dataset and methods. The PTU-IRIS dataset showed GARs of 97.5% for Lacunae, 86.5% for Wolfflin Nodules, 79.8% for Contraction Furrows, and 79.4% for Pigment Spots. Table 1: Comparison of proposed features with the PTU-IRIS dataset DATASETS METHODS GAR(%) PTU-IRIS Lacunae 97.5 Wolfflin Nodules 86.5 Contraction Furrows 79.8 Pigment spots 79.4 CONCLUSION In this paper, the results illustrate the variability in the effectiveness of different biometric features across datasets, highlighting Lacunae as a consistently high-performing feature. The Iris-FE, a deep convolutional neural network, excels in automatically learning hierarchical features, capturing the global spatial relationships within the iris image. By leveraging its deep architecture, Iris-FE extracted robust features from complex, textured regions, enhancing the overall accuracy and efficiency of the system. The local features focus on capturing local gradients and edge directions, which are crucial for recognizing fine details in the iris texture. This enables the system to handle high-dimensional feature spaces by providing a comprehensive feature set that encapsulates both global patterns and local details. By comparing the GAR values, it is possible to better understand the trade-offs and potential of each feature, leading to the development of more robust and reliable biometric authentication systems. Declarations Author Contribution Poovayar priya has contributed to the research findings and methodology, Ezhilarasan contributed to the supervision and evaluation. References K. Nguyen, H. Proenca and F. Alonso-Fernandez, “Deep learning for iris recognition: A survey”, ACM Computing Surveys , vol. 56, no. 9, pp. 1–35, 2024. A. K. Jain, P. Flynn and A. Ross, “Handbook of Biometrics”, Springer Science Business Media , New York, 2007. H. Van de Haarvan, D. Greunen, and D. Pottas, “The characteristics of a biometric”, In proceedings of 2013 Information Security for South Africa, Johannesburg, South Africa , pp. 1–8, Oct. 2013. R.P. Wildes, “Iris recognition: an emerging biometric technology”, In Proceedings of the IEEE conference , vol. 85, no. 9, pp. 1348–1363, 1997. Minaee, S. and Abdolrashidi, A.,” Deepiris: Iris recognition using a deep learning approach”, 2019. Nguyen, K., Fookes, C., Ross, A. and Sridharan, S., “Iris recognition with off-the-shelf CNN features: A deep learning perspective”, IEEE Access , vol. 6 , pp.18848–18855, 2017. Nguyen, K., Proença, H. and Alonso-Fernandez, F.,” Deep learning for iris recognition: A survey”, ACM Computing Surveys , vol. 56 , no. 9, pp.1–35, 2024. I. Pavaloi, C.D. Nita, and L.C. Lazar, “Experiments on iris classification and retrieval with fixed number of SURF keypoints and texture features”, Memoirs of the Scientific Sections of the Romanian Academy , vol. 46, pp. 127–138, Mar. 2023. B. Herbert, “Surf: Speeded up robust features”, Computer vision and image understanding , vol. 110, no. 3, pp.346–359, 2008. A. Baraldi and F. Panniggiani, “An investigation of the textural characteristics associated with gray level cooccurrence matrix statistical parameters”, IEEE Transactions on Geoscience and remote sensing , vol. 33, no. 2, pp.293–304, 1995. F. Ozbilgin, C. Kurnaz, and E. Aydın, “Prediction of coronary artery disease using machine learning techniques with iris analysis”, Diagnostics , vol. 13, no. 6, pp. 1–20, 2023. D.G. Lowe, “Object recognition from local scale-invariant features”, In Proceedings of the seventh IEEE International Conference on computer vision, Kerkyra, Greece, vol. 2, pp. 1150–1157, Sept. 1999. A. Boyd, D. Moreira, D. Kuehlkamp, K. Bowyer and A. Czajka, “Human saliency-driven patch-based matching for interpretable post-mortem iris recognition”, In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, Waikoloa, HI, USA, pp. 701–710, Jan. 2023. Shen, B., Xu, Y., Lu, G. and Zhang, D.,” Detecting iris lacunae based on Gaussian filter”, In Third International Conference on Intelligent Information Hiding and Multimedia Signal Processing (IIH-MSP 2007) , vol. 1, pp. 233–236, IEEE, Nov. 2007. Sindu, M., Thiyaneswaran, B., Anguraj, K. and Yoganathan, N., “Extraction of Iris Crypt, Pigment Spot, and Wolfflin Nodule Biological Feature using Feature Point Selection Algorithms”, International Journal of Recent Technology and Engineering (IJRTE) , vol. 8 , no. 4, pp.2225–2230, 2019. Chua, J., Thakku, S.G., Pham, T.H., Lee, R., Tun, T.A., Nongpiur, M.E., Tan, M.C.L., Wong, T.Y., Quah, J.H.M., Aung, T. and Girard, M.J., “Automated detection of iris furrows and their influence on dynamic iris volume change”, Scientific reports , vol. 7 , no. 1, p.17894, 2017. Khan, A.R., Javed, R., Sadad, T., Bahaj, S.A., Sampedro, G.A. and Abisado, M., “Early pigment spot segmentation and classification from iris cellular image analysis with explainable deep learning and multiclass support vector machine”, Biochemistry and Cell Biology , 2023. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 03 Mar, 2026 Reviewers agreed at journal 27 Feb, 2026 Reviewers invited by journal 26 Dec, 2024 Editor assigned by journal 28 Oct, 2024 Submission checks completed at journal 28 Oct, 2024 First submitted to journal 28 Oct, 2024 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-5345891","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":373366363,"identity":"a0b911da-18db-43db-be2e-fa0f278a3031","order_by":0,"name":"Poovayar Priya M","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABEElEQVRIiWNgGAWjYBACCSA+AKbY2w8+AImwgQgeMCKkhedMsgFDApFaoCwHMwmQFgaoFpxAsv2M4eHCPRaJ/TMY0qoLfxzO45NIYHzwto1BxhyHFmmeHIPDM55JJM643Xjs9oyEw8VsEgnMhnPbGHgsG7BrkWNISzjMc0AiseHOgbTbPAmHE9t4DrBJ8wK1GBzAoYX/GUTL/BsJZsVQLey/8WmRlkg+ANayAaiFGayFvYGNGZ8WyRmPDxyecUDCeOOZM8nSM9LSgVoamyXnnJPAqUXifGLz54IDdbLzjrcf/FxgY504v5n54Ic3ZTb2uLSAADMQOzZAGUDACGJL4FQO02IPY4yCUTAKRsEowAAAXnxdNaoe6FMAAAAASUVORK5CYII=","orcid":"","institution":"Puducherry Technological University","correspondingAuthor":true,"prefix":"","firstName":"Poovayar","middleName":"Priya","lastName":"M","suffix":""},{"id":373366364,"identity":"244f0107-929c-4841-bcc7-bb531404e4b3","order_by":1,"name":"Ezhilarasan M","email":"","orcid":"","institution":"Puducherry Technological University","correspondingAuthor":false,"prefix":"","firstName":"Ezhilarasan","middleName":"","lastName":"M","suffix":""}],"badges":[],"createdAt":"2024-10-28 09:38:28","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5345891/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5345891/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":69356412,"identity":"27c983af-be0f-46d8-8e1c-bb1d8ca2ec1c","added_by":"auto","created_at":"2024-11-19 13:49:52","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":12884,"visible":true,"origin":"","legend":"\u003cp\u003eIris features\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5345891/v1/219a51b696a8cfc1ded860d9.jpg"},{"id":69356413,"identity":"ddf3916a-70de-49ac-94c8-a6f0002a247c","added_by":"auto","created_at":"2024-11-19 13:49:52","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":122874,"visible":true,"origin":"","legend":"\u003cp\u003eLacunae signs in the iris\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5345891/v1/64f7ecb7402fd9b03bdf7128.jpg"},{"id":69355548,"identity":"8b36ab63-5468-42c2-97f8-fb469c182f07","added_by":"auto","created_at":"2024-11-19 13:41:52","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":70869,"visible":true,"origin":"","legend":"\u003cp\u003eWolfflin nodules signs in the iris\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5345891/v1/d2bf59350cf09776ad5659a0.jpg"},{"id":69355553,"identity":"c543c5fc-91c6-4da4-8a26-8b06d8ae4c19","added_by":"auto","created_at":"2024-11-19 13:41:52","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":84412,"visible":true,"origin":"","legend":"\u003cp\u003eContraction furrow signs in the iris\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5345891/v1/1dd7ca9407ef36927151067e.jpg"},{"id":69355552,"identity":"160d2a36-3669-4030-baa3-e1d4cd0b4798","added_by":"auto","created_at":"2024-11-19 13:41:52","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":100347,"visible":true,"origin":"","legend":"\u003cp\u003ePigment Spots signs in the iris\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5345891/v1/8cbe7b43da0c30afb57d7bb0.jpg"},{"id":69355551,"identity":"e64c0c99-bda7-43ad-82b7-8c38bb67e5c3","added_by":"auto","created_at":"2024-11-19 13:41:52","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":47152,"visible":true,"origin":"","legend":"\u003cp\u003eIris-FE Architecture for Feature Extraction\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5345891/v1/a62c2e9362144b641cbe5ca5.jpg"},{"id":69355546,"identity":"ca78f23a-0dfb-4152-9b09-d240eb374152","added_by":"auto","created_at":"2024-11-19 13:41:52","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":38869,"visible":true,"origin":"","legend":"\u003cp\u003eROC curve of proposed features for the PTU-IRIS dataset\u003c/p\u003e","description":"","filename":"7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5345891/v1/83cf06aca8f970bac362b7f1.jpg"},{"id":69355549,"identity":"2e01fabc-cbfe-41c3-8c77-f4fd74190df3","added_by":"auto","created_at":"2024-11-19 13:41:52","extension":"jpg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":72327,"visible":true,"origin":"","legend":"\u003cp\u003eDET curve of proposed features for the PTU-IRIS dataset\u003c/p\u003e","description":"","filename":"8.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5345891/v1/700e513fbf2874ad85375860.jpg"},{"id":69356414,"identity":"65004b14-7f73-44b8-82d1-815e78f68c3b","added_by":"auto","created_at":"2024-11-19 13:49:57","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":958955,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5345891/v1/87b817b8-957b-4971-96e9-955f0f81dddc.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"\u003cp\u003eA Novel Approach to the Extraction of Iris Features Using a Deep Learning Network\u003c/p\u003e","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eThe iris is a sensitive, ring-like structure located between the sclera and the pupil. It plays a crucial role in controlling the size of the pupil, thus managing the amount of light that enters the retina. The iris, which encircles the pupil, exhibits unique features such as furrows, pits, and crypts that are specific to each person and remain consistent throughout life, even in identical twins. Unlike fingerprints, the iris is more resistant to forgery due to the protective sclera. Iris recognition is a sophisticated biometric technique that utilizes the visible iris for authentication. This allows for non-invasive capture of images from distances of 7.5 to 15 cm (3 to 10 inches). The iris\u0026rsquo;s enduring and intricate patterns make it a highly reliable biometric identifier [1].\u003c/p\u003e \u003cp\u003eSince the iris image obtained from the sensor is typically raw and unprocessed, it is not directly used for authentication. Instead, the iris's inherent textures and patterns known for their uniqueness are utilized for identification purposes [2]. The iris is a complex structure with distinctive features such as the pupil, pupillary zone, collarette, ciliary zone, radial furrows, and crypts as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. These features indicate the intricate pattern of the iris, which and its color variations play a crucial role in identification.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe process relies heavily on analyzing the intricate textures of the iris, which contain rich spatial frequency information. Techniques like Gabor filters and wavelet transforms are often used to emphasize the high-frequency details, such as crypts and furrows, that distinguish one iris from another. Multi-resolution feature extraction methods allow for the capture of both fine and coarse iris structures, ensuring that no critical detail is overlooked [3]. This ability to capture complex patterns, combined with the iris's natural protection by the sclera and eyelids, makes iris recognition highly resistant to forgery. Moreover, despite the high-dimensional nature of the data, the feature extraction process efficiently condenses the information into a compact, distinctive representation, enabling rapid and accurate matching against stored templates. This makes iris recognition one of the most secure and precise biometric identification systems [4].\u003c/p\u003e \u003cp\u003eThe use of deep learning algorithms in iris recognition has significantly improved the accuracy and robustness of feature extraction. Traditional feature extraction methods, such as Gabor filters and wavelet transforms, rely on handcrafted features that capture specific textures of the iris. While effective, these methods may not always account for the wide variability in iris images due to noise, lighting, or occlusions. Deep learning, on the other hand, enables automatic learning of feature representations directly from the raw image data, allowing the model to capture more complex and subtle patterns in the iris that traditional methods might miss [5]. Convolutional Neural Networks (CNNs), in particular, have revolutionized iris recognition by automatically learning hierarchical features from the input image. Instead of relying on pre-defined filters, CNNs can learn optimal filters that extract the most relevant features from the iris at multiple levels of abstraction. This results in a more comprehensive and accurate representation of the iris, improving recognition performance even under challenging conditions, such as poor lighting or partial occlusion by eyelids or eyelashes [6].\u003c/p\u003e \u003cp\u003eFurthermore, deep learning models can generalize better to variations in iris images, such as those caused by changes in pupil size, head tilt, or noise in the image. Advanced architectures like Res-Net, VGG-Net, and MobileNetV2, when applied to iris recognition, can capture fine-grained details while being robust to distortions. These networks can also be fine-tuned to enhance specific aspects of iris feature extraction, leading to state-of-the-art accuracy and reduced error rates. Incorporating deep learning algorithms has not only improved the precision of iris recognition systems but also made them more adaptive and scalable. This ability to learn and refine features without manual intervention offers significant advantages over traditional methods, positioning deep learning as a key technology for the future of biometric recognition [7].\u003c/p\u003e"},{"header":"LITERATURE SURVEY","content":"\u003cp\u003eIn iris recognition, existing feature extraction techniques are generally categorized into three main types: texture-based, geometric-based, and macro-based feature extraction. Each of these approaches targets different aspects of the iris, allowing the system to capture a wide range of distinguishing details for accurate identification.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e1. \u0026nbsp; Texture features\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe iris region is divided into the pupillary zone and ciliary zone for feature extraction and encoding. The Daugman rubber sheet model is employed to map the iris texture features from these zones into a feature vector. This model transforms the entire iris region into a rectangular shape by considering the density of the pupillary zone, which is closer to the pupil and less affected by obstructions such as eyelids or lashes. The pupillary zone, with its lower frequency and richer information, is preferred for identification over the ciliary zone.\u003c/p\u003e\n\u003cp\u003eBefore feature extraction, the segmented iris image undergoes normalization to address iris size and lighting variations. This process is essential to counteract the effects of elastic deformation, which can impact matching performance. For iris image classification, texture features [8] are combined with Speeded Up Robust Features (SURF) [9] key points. The Descriptor Number On Limits (DENOL) method generates a fixed number of SURF key points for each image in the dataset by adjusting the Hessian threshold parameter. SURF is known for its computational efficiency and robustness to scale changes, rotation, and illumination, making it suitable for real-time applications and large-scale data processing.\u003c/p\u003e\n\u003cp\u003eSURF operates through a series of Gaussian convolutions to create a scale space for\u003c/p\u003e\n\u003cp\u003emulti-scale feature detection. It identifies interest points by analyzing the Hessian matrix determinants and detects local maxima to ensure stable \u0026amp; repeatable features. Each point of interest is analyzed within a square region, subdivided into smaller areas for descriptor extraction. Haar wavelet responses are computed in both horizontal and vertical directions for each sub-region, creating descriptor vectors that represent the spatial distribution of local image gradients. Comparing these descriptors across images involves measuring the Euclidean distance between vectors to identify matching local structures. Iris feature extraction faces performance issues in certain conditions, particularly with highly textured iris images.\u003c/p\u003e\n\u003cp\u003eThe structure of the texture features is as follows:\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePupillary Zone\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCiliary Zone\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eOuter Edge\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eThe following is the representation of the feature vector stored in the Database:\u003c/p\u003e\n\u003cp\u003ea. Size: 2048 bits (the total length of the feature vector).\u003c/p\u003e\n\u003cp\u003eb. Position: Specific positions within the vector might encode different types of\u003c/p\u003e\n\u003cp\u003etexture information.\u003c/p\u003e\n\u003cp\u003ei. Positions 0\u0026ndash;511: Encodes texture patterns from the pupillary zone.\u003c/p\u003e\n\u003cp\u003eii. Positions 512\u0026ndash;1535: Encodes texture patterns from the ciliary zone.\u003c/p\u003e\n\u003cp\u003eiii. Positions 1536\u0026ndash;2047: Encodes texture patterns from the outer edge.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2. \u0026nbsp; Geometric features\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGeometric traits of the iris are closely linked to its shape which are valuable forbiometric identification. These geometric features can be leveraged at various levelsof identification. While individual features like the pupil\u0026apos;s size might not be highlydistinctive on their own, they become uniquely identifying when combined with othercharacteristics such as the iris\u0026apos;s texture or the collarette. The integration of these geometricelements enhances the overall uniqueness of the iris which makes it a robustfeature for identification purposes.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003ei. Pupil Based Geometric features\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAlthough the pupil is generally defined as circular, it is not always perfectly circular. However, the geometric characteristics still need to be recognized and identify the pupil. The following sections will consider and describe the pupil roundness, size, and smoothness.\u003c/p\u003e\n\u003cp\u003ea. \u0026nbsp; Pupil Roundness\u003c/p\u003e\n\u003cp\u003eSeveral orientation points on the iris can be used to assess the consistency of its diameter, as roundness measurement involves tracing the iris 360o around. Generally, roundness is defined by the degree of the edges and corners of a circle related to the circle\u0026rsquo;s compactness and sphericity.\u003c/p\u003e\n\u003cp\u003eb. \u0026nbsp; Pupil Largeness\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe pupil radius is used to characterize its size. However, the general radius is\u0026nbsp;\u003c/p\u003e\n\u003cp\u003einadequate because the pupil is not a perfect circle. Eight directions on the pupil: 0\u003csup\u003eo\u003c/sup\u003e, 45\u003csup\u003eo\u003c/sup\u003e, 90\u003csup\u003eo\u003c/sup\u003e, 135\u003csup\u003eo\u003c/sup\u003e, 180\u003csup\u003eo\u003c/sup\u003e, 225\u003csup\u003eo\u003c/sup\u003e, 270\u003csup\u003eo\u003c/sup\u003e and 315\u003csup\u003eo\u003c/sup\u003e are used to compute the radius.\u003c/p\u003e\n\u003cp\u003ec. \u0026nbsp; Pupil Smoothness\u003c/p\u003e\n\u003cp\u003eThe curvature can be used to represent the smoothness of the iris by a circle, which is defined as the reciprocal of the radius (1/r), Since the pupil is regarded as a circle, the Pupil Smoothness (PS) is measured using the curvature of the circle.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eii. Collarette-Based Geometric features\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe collarette region, located in the inner part of the iris remains unaffected by externalfactors such as eyelids or lashes and is not influenced by pupil dilation. This areahas a fixed radius range and is concentric with the pupil. Characterized by distinct dotsor radial spokes, the collarette\u0026apos;s geometric features must first be identified within the irisbefore any geometric measurements can be calculated. Its stable location and uniquepattern make it a reliable feature for biometric identification.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eiii. Geometric Features of Iris\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn iris identification, geometric features are crucial as they provide distinct and measurable attributes that enhance the accuracy and reliability of biometric identification systems. Key geometric parameters such as iris proximity and diameter offer valuable insights into the intricate structural details of the iris. The iris diameter, a fundamental metric in the normalization process measures the horizontal distance from the pupil\u0026apos;s centre to the iris\u0026apos;s edge which ensures consistency across different images. Additionally, iris proximity measures the spatial relationships between the iris, pupil and sclera which captures the radial distance from the pupil centre to the iris edge and other defining features. These geometric measurements contribute significantly to distinguishing and verifying individual irises.\u003c/p\u003e\n\u003cp\u003ea. Iris diameter\u003c/p\u003e\n\u003cp\u003eThe average diameter of the iris at eight different locations is used to compute this characteristic. Simply measuring the distance from the iris\u0026rsquo;s center to any point on its outer edge is insufficient, as the iris is not a perfect circle. Its shape can vary, making it important to consider multiple measurements to accurately represent its dimensions. Therefore, the average iris diameter is determined on the pupil at twelve different angles: 0\u003csup\u003eo\u003c/sup\u003e, 15\u003csup\u003eo\u003c/sup\u003e, 30\u003csup\u003eo\u003c/sup\u003e, 45\u003csup\u003eo\u003c/sup\u003e, 60\u003csup\u003eo\u003c/sup\u003e, 75\u003csup\u003eo\u003c/sup\u003e, 90\u003csup\u003eo\u003c/sup\u003e, 105\u003csup\u003eo\u003c/sup\u003e, 120\u003csup\u003eo\u003c/sup\u003e, 135\u003csup\u003eo\u003c/sup\u003e, 150\u003csup\u003eo\u003c/sup\u003e and 165\u003csup\u003eo\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eb. Iris proximity\u003c/p\u003e\n\u003cp\u003eA line that crosses a circle exactly once is called a tangent to it. The term \u0026quot;point tangency\u0026quot; refers to the junction point. At the precise point of tangency, which is a straight angle of 90\u003csup\u003eo\u0026nbsp;\u003c/sup\u003eand perpendicular to the line drawn from the centre, the tangent hits the circle. From the outer point (sclera) to the iris, a tangent is drawn.\u003c/p\u003e\n\u003cp\u003eThe iris image was converted into a rectangular format using integral differential operators and rubber sheet methods, followed by cropping the rectangular region based on the iris map. Feature extraction was carried out using various techniques, including wavelet transform, first-order statistical analysis, Gray-Level Co-Occurrence Matrix (GLCM) [10], and Gray Level Run Length Matrix (GLRLM) [11].\u0026nbsp;Wavelet transform is particularly effective for processing signals and images, enabling multi-resolution analysis that decomposes the image into different frequency components. This approach captures both spatial and frequency information simultaneously which is essential for highlighting geometric features and ensuring spatial localization thus simplifying the detection of localized features critical for distinguishing between different irises. First-order statistical analysis computes basic image statistics such as average, standard deviation, median, and minimum \u0026amp; maximum values. These statistics describe the image\u0026rsquo;s intensity distribution without considering spatial relationships, offering insights into the overall geometric features and intensity variations of the iris.\u003c/p\u003e\n\u003cp\u003eThe GLCM is a statistical method used to examine spatial relationships between pixels in an image. It captures geometric features through metrics such as contrast, correlation, energy, homogeneity, and entropy by analyzing how frequently specific pairs of pixel intensities occur at various distances and angles. GLCM is effective in detecting repetitive directional patterns in the iris which are useful for texture analysis in iris identification. The GLRLM estimates the size of continuous pixels with similar gray levels at given angles. Attributes from GLRLM such as short-run emphasis, long-run emphasis, gray level non-uniformity, run length non-uniformity and run percentage help to determine the roughness and orientation of geometric features. GLRLM is valuable for characterizing the iris as it captures directional patterns involved in geometric features across multiple directions. These systems often struggle to handle high-dimensional feature spaces, which can result in lower accuracy and efficiency.\u003c/p\u003e\n\u003cp\u003eThe structure of the geometric features are as follows:\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp; PS\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;ID\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eIP\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003ea. Pupil Roundness (PR)\u003c/p\u003e\n\u003cp\u003eA single floating-point value representing the roundness of the pupil consists of 32\u003c/p\u003e\n\u003cp\u003ebits.\u003c/p\u003e\n\u003cp\u003eb. Pupil Largeness (PL)\u003c/p\u003e\n\u003cp\u003eAn array of 8 floating-point values with 256 bits representing the radius at different\u003c/p\u003e\n\u003cp\u003eangles (0\u0026deg;, 45\u0026deg;, 90\u0026deg;, 135\u0026deg;, 180\u0026deg;, 225\u0026deg;, 270\u0026deg;, 315\u0026deg;).\u003c/p\u003e\n\u003cp\u003ec. Pupil Smoothness (PS)\u003c/p\u003e\n\u003cp\u003eAn array of 8 floating-point values with 256 bits representing the smoothness at \u0026nbsp; \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDifferent angles which is computed as the curvature (1/radius).\u003c/p\u003e\n\u003cp\u003ed. Collarette Radius (CR)\u003c/p\u003e\n\u003cp\u003eAn array of 12 floating-point values with 384 bits representing the radius of the\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCollarette at different angles (0\u0026deg;, 15\u0026deg;, 30\u0026deg;, ..., 165\u0026deg;).\u003c/p\u003e\n\u003cp\u003ee. Iris Diameter (ID)\u003c/p\u003e\n\u003cp\u003eAn array of 12 floating-point values with 384 bits representing the average diameter\u003c/p\u003e\n\u003cp\u003eof the iris at different angles (0\u0026deg;, 15\u0026deg;, 30\u0026deg;, ..., 165\u0026deg;).\u003c/p\u003e\n\u003cp\u003ef. Iris Proximity (IP)\u003c/p\u003e\n\u003cp\u003eAn array of 8 floating-point values with 256 bits representing the proximity between\u003c/p\u003e\n\u003cp\u003ethe iris and the sclera at different angles.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3. \u0026nbsp; Macro features\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe visible features that constitute the structural components of the iris and render it unique for identification are referred to as macro-level features. These macro characteristics include noticeable discontinuities in the iris\u0026rsquo;s size, color, texture, and intensity. The iris comprises numerous intricate and distinctive patterns, which can be leveraged for identification. In this context, radial furrows and crypts are particularly significant for distinguishing individuals as detailed in the following sections:\u003c/p\u003e\n\u003cp\u003ei. Crypts\u003c/p\u003e\n\u003cp\u003eThe crypts are minute yet significant details of the iris, appear as pit-like depressions or extra apertures found around the outer edge of the ciliary region and near the collarette. These crypts facilitate the flow of aqueous humor from the stroma and deeper iris tissues. Typically, crypts are located closer to the collarette and display a darker hue compared to the surrounding iris areas. The structure of the crypts:\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSize\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;Position\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eIntensity\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003ea. Size: A floating-point value representing the average size of the crypts\u003c/p\u003e\n\u003cp\u003ewith 32 bits.\u003c/p\u003e\n\u003cp\u003eb. Position: An array of 8 floating-point values with 256 bits representing\u003c/p\u003e\n\u003cp\u003ethe positions of the crypts relative to the collarette in polar coordinates (r,\u003c/p\u003e\n\u003cp\u003e\u0026theta;).\u003c/p\u003e\n\u003cp\u003ec. Intensity: A floating-point value representing the average intensity of the crypts\u003c/p\u003e\n\u003cp\u003ecompared to the surrounding iris tissue with 32 bits.\u003c/p\u003e\n\u003cp\u003eii. Radial Furrows\u003c/p\u003e\n\u003cp\u003eRadial furrows extend outward in all directions from the centre of the pupil, potentially starting near the pupil and extending through the collarette. These fissures in the anterior layer allow for simulation of the iris with loose tissue sometimes protruding outward. This feature is height-dependent due to the outward bulging. Despite their variable appearance, the relative positions of radial furrows remain stable. The structure of the radial furrows:\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLength\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePosition\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eIntensity\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003ea. Length: An array of 8 floating-point values with 256 bits representing\u003c/p\u003e\n\u003cp\u003ethe length of radial furrows measured from the pupil outward.\u003c/p\u003e\n\u003cp\u003eb. Position: An array of 8 floating-point values with 256 bits representing\u003c/p\u003e\n\u003cp\u003ethe relative position of radial furrows along the circumference of the iris.\u003c/p\u003e\n\u003cp\u003ec. Intensity: A floating-point value with 32 bits representing the intensity\u003c/p\u003e\n\u003cp\u003evariation within the radial furrows compared to the surrounding iris tissue.\u003c/p\u003e\n\u003cp\u003eThe use of the Scale-Invariant Feature Transform (SIFT) [12] Scale-Invariant Feature Transform algorithm for detecting key points and performing iris matching is highlighted in [13]. SIFT is particularly advantageous for iris identification due to its robustness in detecting distinctive features that remain invariant to scale, rotation and illumination changes. This characteristic makes SIFT well-suited for extracting and identifying fine patterns in the iris such as crypts and radial furrows. SIFT operates by constructing a scale space using the Difference of Gaussians (DoG) approach to detect potential key points. It then localises these key points accurately and assigns orientations based on the local gradient direction which ensures rotation invariance. The final step involves generating descriptors by computing gradient magnitudes and orientations within a local region around each key point, capturing unique texture information.\u003c/p\u003e\n\u003cp\u003eIn iris identification, SIFT can effectively identify key points like crypts are small, irregular dark spaces resulting from iris tissue separations. Radial furrows are lines extending from the pupil. The scale and rotation invariance of SIFT combined with its robustness to illumination changes, enhances the accuracy and reliability of iris identification systems. By providing consistent and detailed feature information regardless of variations in eye orientation, image resolution, or lighting conditions. Moreover, many iris feature extraction algorithms focus on the local neighborhood of each pixel and fail to capture global spatial relationships within the iris image.\u003c/p\u003e"},{"header":"MATERIALS AND METHODS","content":"\u003cp\u003e\u003cstrong\u003eDatasets:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Puducherry Technological University (PTU)-IRIS dataset is used in experiments\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003ewith the\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eproposed work. Two sessions were used to gather the iris images utilizing a Mobile-EyesTM\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003esensor, a product of the L-1 Identify solution. There are 5000 participants in the dataset; 4000 were\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eused for training and 1000 for testing from faculty members and students of the University. The\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eindividuals were asked directly if they wore contact lenses during the process of taking their\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eimages and this was noted. The images were taken with near-infrared (NIR) light. It records\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eSVGA images with a width of 752 x 480 at a rate of 30 frames per second. Due to the dual-iris\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eenrollment feature of the gadget, simultaneous auto-capture of the left and right eyes is possible.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eIt can function in both dimly lit and brilliantly lit areas. The scanned image has a resolution of\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e640 x 480 and is saved in the PNG file format.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eProposed Work:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFeature extraction is a crucial part of an iris identification system. In the existing systems, there are three iris features such as textural features for iris patterns, geometric features for iris shape, pupil \u0026amp; collarette, and macro features for visual information. However, the existing features suffer from performance issues in certain conditions, particularly with highly textured iris images. These features often struggle to handle high-dimensional feature spaces which can result in lower accuracy and efficiency. Additionally, many iris feature extraction algorithms focus on the local neighborhood of each pixel and fail to capture global spatial relationships within the iris image. The iris surface features proposed in this paper address all these issues in the existing systems.\u003c/p\u003e\n\u003cp\u003eIn iris identification, individual differentiation depends on a thorough examination of iris surface features. These surface features include lacunae,\u0026nbsp;Wolfflin nodules, Contraction furrows, and pigment spots. These distinctive iris textures and patterns are important for biometric identification. These features are compared with the existing features for better identification results. The characteristics of iris surface features are discussed in the following subsections.\u003c/p\u003e\n\u003cp\u003ei. \u0026nbsp; \u0026nbsp;Lacunae\u003c/p\u003e\n\u003cp\u003eLacunae are distinct clustered lesions found in the iris and play a significant role in iris identification systems. During the identification process, these lacunae are detected and modeled as clustered shapes with parameters like size, orientation, and count, which are extracted as unique features of the individual\u0026apos;s iris. Due to their unique patterns and distribution, lacunae as shown in Figure 2 contribute to creating a highly individualized biometric template. If a closed lesion is found during iris diagnosis, it is strongly advised that the patient be instructed to have a comprehensive checkup. The count of each Lacunae measure is extracted as features [14].\u003c/p\u003e\n\u003cp\u003eFigure 2: Lacunae signs in the iris\u003c/p\u003e\n\u003cp\u003eii. Wolfflin Nodules\u003c/p\u003e\n\u003cp\u003eIn the iris identification system, Wolfflin nodules play a critical role by providing unique biometric features for identification. During the identification process, Wolfflin nodules as shown in Figure 3 characterized by their open-ended, circular, or near-circular raised areas which are detected in the iris image. These nodules are indicative of ongoing metabolic processes and can reveal information about the state of associated organs. The count and characteristics of each Wolfflin nodule are extracted as distinct features of the iris. These features are then incorporated into the biometric template which represents the unique characteristics of an individual\u0026apos;s iris. During identification, the extracted features from a newly captured iris image are compared against the stored templates in the database [15].\u003c/p\u003e\n\u003cp\u003eFigure 3: Wolfflin nodules signs in the iris\u003c/p\u003e\n\u003cp\u003eiii. Contraction Furrows\u003c/p\u003e\n\u003cp\u003eContraction rings show how the neurological and muscular systems are related. Irritative stress causes long-lasting contractive spasms that obstruct the regular flow of fluids that bring waste products and nutrients to the muscles. The ring itself forms a valley or mountain ridge and is frequently referred to as a contraction furrow as shown in Figure 4. There are occasions when the individual iris fibers exhibit a zigzag pattern in addition to this ridge. The ridge appears to be a valley as the nerve ring becomes deeper. One or more segments of curving furrows that follow the circumference of the iris and range in color from bright white to dark reveal nerve rings also known as neurovascular cramp rings. They display a level of stress and anxiety. Nerve rings indicate tension and a desire for rest. This may come from physical or mental areas. Contraction furrows are modeled as concentric circles with varying radii. The count of each furrow is taken as a feature of Contraction furrows [16].\u003c/p\u003e\n\u003cp\u003eFigure 4: Contraction furrow signs in the iris\u003c/p\u003e\n\u003cp\u003eiv. Pigment Spots\u003c/p\u003e\n\u003cp\u003ePsora, often known as drug spots, are irregular pigment marks that show the buildup of drugs or products of malfunctioning metabolism as shown in Figure 5. When inherited, these characteristics are regarded as miasms or inherited stains from ancestor illnesses. Pigment spots are modeled as small circular regions with intensity or color variations. The count of each spot is considered a feature of pigment spots [17].\u003c/p\u003e\n\u003cp\u003eFigure 5: Pigment Spots signs in the iris\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eIris-FE Architecture for Feature Extraction\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe Iris-FE Architecture for feature extraction as shown in Figure 6 extracts these features and combines the global details with the local details. Convolutional neural network (CNN) architecture, Iris-FE is renowned for its efficiency and simplicity. It comprises several small (3x3) receptive field convolutional layers, followed by fully connected layers. A feature descriptor called Iris Surface Local Features (ISLF) measures how gradient orientations are distributed in specific areas of an image. The Iris-FE extracts high-level information and local details separately, combining both, it extracts both global details as well as local details.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe proposed algorithm takes a normalized and enhanced grayscale iris image (Igray) as input. After that, the image is scaled to 224x224 pixels in resolution to make it compatible with the Iris-FE model which is a standard input size. In order to comply with the limitations of the model and preserve the spatial integrity of the iris features, this scaling is essential. The image is resized and then goes through a preprocessing stage where its dimensions are increased and further normalized to satisfy the deep learning model\u0026apos;s input specifications.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe feature extraction process is divided into four key features, each targeting specific characteristics of the iris. The contours are used to delineate the boundaries of various iris surface features such as lacunae, Wolfflin nodules, contraction furrows, and pigment spots. These contours are typically found by applying edge detection algorithms to a binary image which simplifies the process by reducing the image to its most basic structure. Lacunae features are extracted by detecting contours within the binary version of the iris image (Ibinary). The algorithm identifies and counts closed contours as Lacunae. Wolfflin nodules are quantified by counting the total number of contours detected providing a measure of these raised areas on the iris surface. Contraction furrows, which are large and have deep curvature are identified by detecting contours with an area exceeding a predefined threshold (\u0026lsquo;T\u0026rsquo;). Finally, pigment spots are extracted using a Blob detection technique, where the algorithm counts the number of distinct blobs or key points detected in the binary image.\u003c/p\u003e\n\u003cp\u003eFigure 6: Iris-FE Architecture for Feature Extraction\u003c/p\u003e\n\u003cp\u003eOnce these features are extracted, the Lacunae and Wolfflin nodules features are fed into the global feature model to extract deep features (F\u003csub\u003eISGF\u003c/sub\u003e). ISGF is particularly effective in capturing hierarchical and spatially complex patterns making it suitable for analyzing the intricate textures of the iris. Simultaneously, the features of the contraction furrows and pigment spots are processed through the local feature model to extract local features (F\u003csub\u003eISLF\u003c/sub\u003e). This model excels at capturing edge information and local textures, which are vital for detailed iris analysis. The global and local features obtained from these models are combined as the final output providing a comprehensive feature set for further analysis or classification tasks in iris identification systems.\u003c/p\u003e\n\u003cp\u003eThe extraction of features using Iris-FE is as follows:\u003c/p\u003e\n\u003cp\u003eAlgorithm: Iris-FE\u003c/p\u003e\n\u003cp\u003eStep 1: Input the normalized iris image.\u003c/p\u003e\n\u003cp\u003eStep 2: Resize the input image to 224x224 size resolution.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eI\u003csub\u003eresized\u003c/sub\u003e = resize (I\u003csub\u003egray\u003c/sub\u003e, (224,224))\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eStep 3: Expand dimensions and preprocess the image for the Iris-FE.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eI\u003csub\u003epreprocessed\u003c/sub\u003e = preprocessInput (expandDims(I\u003csub\u003eresized\u003c/sub\u003e,0))\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eStep 4:\u003c/p\u003e\n\u003cp\u003e(i) Using contours, the Lacunae features are extracted from the iris\u003c/p\u003e\n\u003cp\u003eSearch the contours in the binary image:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eC = findContours(Ibinary)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eCount the number of closed contours:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eL =\u0026nbsp;\u003c/em\u003e\u003cem\u003e\u0026Sigma;\u003c/em\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003cem\u003e\u0026delta;\u003c/em\u003e\u003cem\u003e(isContourConvex(c))\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u0026nbsp;\u003c/em\u003e\u003cem\u003ec\u003c/em\u003e\u003cem\u003e\u0026isin;\u003c/em\u003e\u003cem\u003eC \u0026nbsp;\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003ewhere L is the Lacunae count, and\u0026nbsp;\u0026delta;\u0026nbsp;is an indicator function that returns 1 if the contour c is closed, and 0 otherwise.\u003c/p\u003e\n\u003cp\u003e(ii) Extract the Wolfflin nodules from the iris using contours. Count all the contours in the binary image.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eW = |C|\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003ewhere |C| is the total number of contours.\u003c/p\u003e\n\u003cp\u003e(iii) Extract the Contraction furrows from the iris using large contour detection. It is defined as:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eF =\u0026nbsp;\u003c/em\u003e\u003cem\u003e\u0026Sigma;\u003c/em\u003e\u003cem\u003ec\u003c/em\u003e\u003cem\u003e\u0026isin;\u003c/em\u003e\u003cem\u003eC\u003c/em\u003e\u003cem\u003e\u0026delta;\u003c/em\u003e\u003cem\u003e(area(c) \u0026gt; T)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003ewhere T is the predefined threshold.\u003c/p\u003e\n\u003cp\u003e(iv) Extract the pigment spots from the iris using a Blob detector. Apply Blob detection to the binary image:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eK = blobDetector(I\u003csub\u003ebinary\u003c/sub\u003e)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eCount the number of key points detected:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eP = |K|\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003ewhere |K| is the number of detected key points.\u003c/p\u003e\n\u003cp\u003eStep 5: Pass all the deep features to the Iris Surface Global Features Iris Surface Global \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFeatures (ISGF) model as:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eF\u003csub\u003eISGF\u003c/sub\u003e = ISGF(L,W)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eStep 6: Pass all the local features to the Iris Surface Local Features (ISLF) model as:\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eF\u003csub\u003eISLF\u003c/sub\u003e = ISLF(F,P)\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe \u0026lsquo;resize\u0026rsquo; function allows a resolution of 224x224 pixels. The reason for resizing is to ensure that the image matches the input size requirement for the Iris-FE model. These models typically expect a fixed input size, so resizing standardizes the input data. The expandDims\u0026rsquo; function represents the model\u0026apos;s expected inputs in the form of \u0026nbsp;batch_size, height, width, and channels.\u003c/p\u003e\n\u003cp\u003eThe structure of different feature vectors is stored in the database as follows:\u003c/p\u003e\n\u003cp\u003ei) \u0026nbsp; \u0026nbsp;The structure of the Lacunae Feature Vector (LFV) is given as follows:\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSize\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePosition\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;Density\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eDistribution\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003ea. Size: Represents the size of each lacuna (e.g., area or diameter).\u003c/p\u003e\n\u003cp\u003eb. Position: The (x, y) coordinates of the lacuna relative to a reference point in the normalized iris image.\u003c/p\u003e\n\u003cp\u003ec. Shape: Describes the shape of the lacuna, which could be \u0026quot;round\u0026quot;, \u0026quot;oval\u0026quot;, \u0026quot;or\u0026nbsp;irregular\u0026quot; etc.\u003c/p\u003e\n\u003cp\u003ed. Distribution: Describes how the lacunae are distributed within the iris, such as \u0026quot;clustered\u0026quot;, \u0026quot;isolated\u0026quot; or \u0026quot;dispersed\u0026quot;.\u003c/p\u003e\n\u003cp\u003eThe total size required for each LFV is:\u003c/p\u003e\n\u003cp\u003eTotal size = Size + Position + Density + Distribution= 4 bytes + 8 bytes + 4 bytes + 1 byte=17 bytes.\u003c/p\u003e\n\u003cp\u003eii) The structure of the Wolfflin nodules Feature Vector (WFV) is given as follows:\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSize\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;Position\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eShape\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003ea. Size: Represents the size of each nodule (e.g., area or diameter).\u003c/p\u003e\n\u003cp\u003eb. Position: The (x, y) coordinates of the nodule relative to a reference point in the normalized iris image.\u003c/p\u003e\n\u003cp\u003ec. Shape: Describes the shape of the nodule, which could be \u0026quot;round\u0026quot;, \u0026quot;oval\u0026quot; \u0026quot;irregular\u0026quot;\u0026nbsp;etc.\u003c/p\u003e\n\u003cp\u003eThe total size required for each WFV is:\u003c/p\u003e\n\u003cp\u003eTotal size = Size + Position + Shape = 4 bytes + 8 bytes + 1 byte=13 bytes\u003c/p\u003e\n\u003cp\u003eiii) The structure of the Contraction Furrows Feature Vector (CFV) is given as follows:\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eCurvature\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;Depth\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLength\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePosition\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003ea. Curvature: This represents the curvature of each furrow, which may vary\u003c/p\u003e\n\u003cp\u003edepending on its location within the iris.\u003c/p\u003e\n\u003cp\u003eb. Depth: Indicates the depth of each furrow\u003c/p\u003e\n\u003cp\u003ec. Length: Measures the length of the furrow, which can be important for distinguishing\u003c/p\u003e\n\u003cp\u003ebetween different furrows.\u003c/p\u003e\n\u003cp\u003ed. Position: The (x, y) coordinates of the furrow relative to a reference point in\u003c/p\u003e\n\u003cp\u003ethe normalized iris image.\u003c/p\u003e\n\u003cp\u003eThe total size required for each CFV is:\u003c/p\u003e\n\u003cp\u003eTotal size=Curvature + Depth + Length+ Position=4 bytes+4 bytes +4 bytes+8 bytes=20 bytes.\u003c/p\u003e\n\u003cp\u003eiv) The structure of the Pigment Spots Feature Vector (PFV) is given as follows:\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSize\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePosition\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eShape\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003ea. Size: Represents the size of each spot (e.g., area or diameter).\u003c/p\u003e\n\u003cp\u003eb. Position: The (x, y) coordinates of the spot relative to a reference point in the normalized iris image.\u003c/p\u003e\n\u003cp\u003ec. Shape: Describes the shape of the spot, which could be \u0026quot;round\u0026quot;, \u0026quot;oval\u0026quot;, \u0026quot;irregular\u0026quot; etc.\u003c/p\u003e\n\u003cp\u003eThe total size required for each PFV is:\u003c/p\u003e\n\u003cp\u003eTotal size = Size + Position + Shape = 4 bytes + 8 bytes + 1 byte=13 bytes\u003c/p\u003e"},{"header":"EXPERIMENTAL RESULTS AND DISCUSSION","content":"\u003cp\u003eFigure 7 shows the accuracy of the acceptance rate of proposed features for the PTU-IRIS dataset. The computation of the distancesbetween biometric pairs was compared to evaluate the acceptance rates at various levels. The ROC curve also shows the trade-offs betweensecurity (low FAR) and usability (high GAR) for the MMU1-IRIS dataset. From the ROCcurve, it is inferred that the GAR values of Lacunae, Wolfflin nodules, Contraction furrows, andpigment spots are 97.5%, 86.5%, 79.8%, and 79.4% respectively. The visualization of this ROCcurve shows that the Lacunae feature outperformed others.\u003c/p\u003e\n\u003cp\u003eFigure 7: ROC curve of proposed features for the PTU-IRIS dataset\u003c/p\u003e\n\u003cp\u003eFigure 8 examines the DET curves for four distinct features of the PTU-IRIS dataset: Lacunae, Wolfflin nodules, Contraction furrows, and Pigment spots. The Error rates for Lacunae, Wolfflin nodules, Contraction furrows, and Pigment spots are 0.025%, 0.05 %, 0.08%, and 0.1% respectively. From this, Lacunae has a lower error rate than other features.\u003c/p\u003e\n\u003cp\u003eFigure 8: DET curve of proposed features for the PTU-IRIS dataset\u003c/p\u003e\n\u003cp\u003eTable 1 systematically evaluates the performance of various biometric authentication methods using the\u0026nbsp;PTU-IRIS dataset focusing on the GAR as a primary metric. The dataset was\u0026nbsp;examined using four distinct biometric features: Lacunae, Wolfflin Nodules, Contraction Furrows, and Pigment Spots. The GAR, which measures the percentage of genuine pairs correctly accepted by the system, varied across the dataset and methods. The PTU-IRIS dataset showed GARs of 97.5% for Lacunae, 86.5% for Wolfflin Nodules, 79.8% for Contraction Furrows, and 79.4% for Pigment Spots.\u003c/p\u003e\n\u003cp\u003eTable 1: Comparison of proposed features with the\u0026nbsp;PTU-IRIS dataset\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDATASETS\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eMETHODS\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eGAR(%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"4\" valign=\"top\"\u003e\n \u003cp\u003ePTU-IRIS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eLacunae\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e97.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eWolfflin Nodules\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e86.5\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eContraction Furrows\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e79.8\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003ePigment spots\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e79.4\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e"},{"header":"CONCLUSION","content":"\u003cp\u003eIn this paper, the results illustrate the variability in the effectiveness of different biometric features across datasets, highlighting Lacunae as a consistently high-performing feature. The Iris-FE, a deep convolutional neural network, excels in automatically learning hierarchical features, capturing the global spatial relationships within the iris image. By leveraging its deep architecture, Iris-FE extracted robust features from complex, textured regions, enhancing the overall accuracy and efficiency of the system. The local features focus on capturing local gradients and edge directions, which are crucial for recognizing fine details in the iris texture. This enables the system to handle high-dimensional feature spaces by providing a comprehensive feature set that encapsulates both global patterns and local details. By comparing the GAR values, it is possible to better understand the trade-offs and potential of each feature, leading to the development of more robust and reliable biometric authentication systems.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003ePoovayar priya has contributed to the research findings and methodology, Ezhilarasan contributed to the supervision and evaluation.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eK. Nguyen, H. Proenca and F. Alonso-Fernandez, \u0026ldquo;Deep learning for iris recognition: A survey\u0026rdquo;, \u003cem\u003eACM Computing Surveys\u003c/em\u003e, vol. 56, no. 9, pp. 1\u0026ndash;35, 2024.\u003c/li\u003e\n\u003cli\u003eA. K. Jain, P. Flynn and A. Ross, \u0026ldquo;Handbook of Biometrics\u0026rdquo;, \u003cem\u003eSpringer Science Business Media\u003c/em\u003e, New York, 2007.\u003c/li\u003e\n\u003cli\u003eH. Van de Haarvan, D. Greunen, and D. Pottas, \u0026ldquo;The characteristics of a biometric\u0026rdquo;, \u003cem\u003eIn proceedings of 2013 Information Security for South Africa, Johannesburg, South Africa\u003c/em\u003e, pp. 1\u0026ndash;8, Oct. 2013.\u003c/li\u003e\n\u003cli\u003eR.P. Wildes, \u0026ldquo;Iris recognition: an emerging biometric technology\u0026rdquo;, \u003cem\u003eIn Proceedings of the IEEE conference\u003c/em\u003e, vol. 85, no. 9, pp. 1348\u0026ndash;1363, 1997.\u003c/li\u003e\n\u003cli\u003eMinaee, S. and Abdolrashidi, A.,\u0026rdquo; Deepiris: Iris recognition using a deep learning approach\u0026rdquo;, 2019.\u003c/li\u003e\n\u003cli\u003eNguyen, K., Fookes, C., Ross, A. and Sridharan, S., \u0026ldquo;Iris recognition with off-the-shelf CNN features: A deep learning perspective\u0026rdquo;, \u003cem\u003eIEEE Access\u003c/em\u003e, vol. \u003cem\u003e6\u003c/em\u003e, pp.18848\u0026ndash;18855, 2017.\u003c/li\u003e\n\u003cli\u003eNguyen, K., Proen\u0026ccedil;a, H. and Alonso-Fernandez, F.,\u0026rdquo; Deep learning for iris recognition: A survey\u0026rdquo;, \u003cem\u003eACM Computing Surveys\u003c/em\u003e, vol. \u003cem\u003e56\u003c/em\u003e, no. 9, pp.1\u0026ndash;35, 2024.\u003c/li\u003e\n\u003cli\u003eI. Pavaloi, C.D. Nita, and L.C. Lazar, \u0026ldquo;Experiments on iris classification and retrieval with fixed number of SURF keypoints and texture features\u0026rdquo;, \u003cem\u003eMemoirs of the Scientific Sections of the Romanian Academy\u003c/em\u003e, vol. 46, pp. 127\u0026ndash;138, Mar. 2023.\u003c/li\u003e\n\u003cli\u003eB. Herbert, \u0026ldquo;Surf: Speeded up robust features\u0026rdquo;, \u003cem\u003eComputer vision and image understanding\u003c/em\u003e, vol. 110, no. 3, pp.346\u0026ndash;359, 2008.\u003c/li\u003e\n\u003cli\u003eA. Baraldi and F. Panniggiani, \u0026ldquo;An investigation of the textural characteristics associated with gray level cooccurrence matrix statistical parameters\u0026rdquo;, \u003cem\u003eIEEE Transactions on Geoscience and remote sensing\u003c/em\u003e, vol. 33, no. 2, pp.293\u0026ndash;304, 1995.\u003c/li\u003e\n\u003cli\u003eF. Ozbilgin, C. Kurnaz, and E. Aydın, \u0026ldquo;Prediction of coronary artery disease using machine learning techniques with iris analysis\u0026rdquo;, \u003cem\u003eDiagnostics\u003c/em\u003e, vol. 13, no. 6, pp. 1\u0026ndash;20, 2023.\u003c/li\u003e\n\u003cli\u003eD.G. Lowe, \u0026ldquo;Object recognition from local scale-invariant features\u0026rdquo;, In Proceedings of the seventh IEEE International Conference on computer vision, Kerkyra, Greece, vol. 2, pp. 1150\u0026ndash;1157, Sept. 1999.\u003c/li\u003e\n\u003cli\u003eA. Boyd, D. Moreira, D. Kuehlkamp, K. Bowyer and A. Czajka, \u0026ldquo;Human saliency-driven patch-based matching for interpretable post-mortem iris recognition\u0026rdquo;, In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, Waikoloa, HI, USA, pp. 701\u0026ndash;710, Jan. 2023.\u003c/li\u003e\n\u003cli\u003eShen, B., Xu, Y., Lu, G. and Zhang, D.,\u0026rdquo; Detecting iris lacunae based on Gaussian filter\u0026rdquo;, In \u003cem\u003eThird International Conference on Intelligent Information Hiding and Multimedia Signal Processing (IIH-MSP 2007)\u003c/em\u003e, vol. 1, pp. 233\u0026ndash;236, IEEE, Nov. 2007.\u003c/li\u003e\n\u003cli\u003eSindu, M., Thiyaneswaran, B., Anguraj, K. and Yoganathan, N., \u0026ldquo;Extraction of Iris Crypt, Pigment Spot, and Wolfflin Nodule Biological Feature using Feature Point Selection Algorithms\u0026rdquo;, \u003cem\u003eInternational Journal of Recent Technology and Engineering (IJRTE)\u003c/em\u003e, vol. \u003cem\u003e8\u003c/em\u003e, no. 4, pp.2225\u0026ndash;2230, 2019.\u003c/li\u003e\n\u003cli\u003eChua, J., Thakku, S.G., Pham, T.H., Lee, R., Tun, T.A., Nongpiur, M.E., Tan, M.C.L., Wong, T.Y., Quah, J.H.M., Aung, T. and Girard, M.J., \u0026ldquo;Automated detection of iris furrows and their influence on dynamic iris volume change\u0026rdquo;, \u003cem\u003eScientific reports\u003c/em\u003e, vol. \u003cem\u003e7\u003c/em\u003e, no. 1, p.17894, 2017.\u003c/li\u003e\n\u003cli\u003eKhan, A.R., Javed, R., Sadad, T., Bahaj, S.A., Sampedro, G.A. and Abisado, M., \u0026ldquo;Early pigment spot segmentation and classification from iris cellular image analysis with explainable deep learning and multiclass support vector machine\u0026rdquo;, \u003cem\u003eBiochemistry and Cell Biology\u003c/em\u003e, 2023.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"signal-image-and-video-processing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"sivp","sideBox":"Learn more about [Signal, Image and Video Processing](http://link.springer.com/journal/11760)","snPcode":"11760","submissionUrl":"https://submission.nature.com/new-submission/11760/3","title":"Signal, Image and Video Processing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Iris Recognition, Feature extraction, Iris surface features, CNN, Deep learning, Iris Surface Global Features, Iris Surface Local Features","lastPublishedDoi":"10.21203/rs.3.rs-5345891/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5345891/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIris recognition is a powerful biometric identification technique due to the uniqueness and complexity of iris patterns. However, existing methods often struggle to capture both global and local iris features, which are essential for accurate and reliable recognition. Specifically, details like lacunae, Wolfflin nodules, contraction furrows, and pigment spots hold significant information about an individual's identity, yet accurately extracting and integrating these features remains challenging. This study introduces the Iris-FE Architecture, a convolutional neural network (CNN)-based solution designed to address these challenges by extracting both global and local iris features in a unified framework. The purpose of this study is to develop and evaluate a CNN architecture, Iris-FE, that captures and combines global and local iris features for more accurate and comprehensive iris recognition. By leveraging advanced feature extraction techniques, this research aims to improve the reliability and performance of iris recognition systems and enhance their practical applications. The proposed Iris-FE Architecture uses Global features (lacunae and Wolfflin nodules) that are fed into the Iris Surface Global Features (ISGF) model, while local features (contraction furrows and pigment spots) are processed through the Iris Surface Local Features (ISLF) model. The outputs of ISGF and ISLF are combined to form a comprehensive feature set, enhancing the system's ability to capture hierarchical patterns and fine-grained details. The Iris-FE Architecture successfully demonstrated its capability to accurately extract and integrate both global and local iris features. Testing on the iris image dataset showed a notable improvement in identification accuracy and robustness compared to traditional feature extraction methods. The model consistently outperformed baseline approaches, particularly in complex images where both global textures and intricate local features were crucial for accurate recognition.\u003c/p\u003e","manuscriptTitle":"A Novel Approach to the Extraction of Iris Features Using a Deep Learning Network","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-11-19 13:41:47","doi":"10.21203/rs.3.rs-5345891/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-03-03T07:26:40+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"315195036280561625746845251648766824723","date":"2026-02-27T10:07:07+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-12-27T03:52:05+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-10-28T16:38:11+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-10-28T16:33:46+00:00","index":"","fulltext":""},{"type":"submitted","content":"Signal, Image and Video Processing","date":"2024-10-28T09:35:26+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"signal-image-and-video-processing","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"sivp","sideBox":"Learn more about [Signal, Image and Video Processing](http://link.springer.com/journal/11760)","snPcode":"11760","submissionUrl":"https://submission.nature.com/new-submission/11760/3","title":"Signal, Image and Video Processing","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"a512da43-edaf-4ec3-82da-932c2884814f","owner":[],"postedDate":"November 19th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2024-11-19T13:41:48+00:00","versionOfRecord":[],"versionCreatedAt":"2024-11-19 13:41:47","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5345891","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5345891","identity":"rs-5345891","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}