A Deep Siamese ResNet-50 Framework with Triplet loss for High-Precision Face Verification

A Deep Siamese ResNet-50 Framework with Triplet loss for High-Precision Face Verification | 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 Deep Siamese ResNet-50 Framework with Triplet loss for High-Precision Face Verification Phan Thi Huong, Huynh Cao Tuan, Nguyen Minh Son, Tran Tay, Thanh Q. Nguyen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8414686/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract This study proposes a deep learning framework for image verification that integrates a Siamese ResNet-50 architecture with triplet loss to enhance feature discrimination, particularly for facial recognition tasks. The model leverages the powerful residual connections of ResNet-50 for robust feature extraction and replaces the conventional contrastive loss with triplet loss to optimize inter-class and intra-class distances in the learned embedding space. Advanced training strategies, including the Adam optimizer, Cosine Annealing learning rate scheduling, and weight decay regularization, are employed to stabilize convergence and improve generalization. The proposed model is evaluated on the challenging Labelled Faces in the Wild (LFW) dataset, achieving 88.33% accuracy, an F1-score of 0.8828, and an AUC-ROC of 0.96. These results outperform baseline architectures such as Siamese VGG16, ResNet-34, and ConvNextTiny, while maintaining a favorable computational complexity. Additional analyses including ROC curves, mean Average Precision (mAP), inference time, and calibration performance demonstrate the model’s superior balance between accuracy, speed, and deployment readiness. These findings highlight the model's strong potential for practical biometric authentication systems and few-shot learning applications. Image verification Face recognition Siamese network ResNet-50 Triplet loss Deep metric learning Biometric authentication Feature embedding Few-shot learning Model calibration Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted 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-8414686","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":594835776,"identity":"c665d168-392d-4d7f-941a-7ba4c37f7bf4","order_by":0,"name":"Phan Thi Huong","email":"","orcid":"","institution":"Lac Hong University","correspondingAuthor":false,"prefix":"","firstName":"Phan","middleName":"Thi","lastName":"Huong","suffix":""},{"id":594835777,"identity":"03af258f-7de2-42e3-86cd-3f8763dafa95","order_by":1,"name":"Huynh Cao Tuan","email":"","orcid":"","institution":"Lac Hong University","correspondingAuthor":false,"prefix":"","firstName":"Huynh","middleName":"Cao","lastName":"Tuan","suffix":""},{"id":594835778,"identity":"be6aef12-2bbe-4298-8bc6-98237d1a5482","order_by":2,"name":"Nguyen Minh Son","email":"","orcid":"","institution":"Lac Hong University","correspondingAuthor":false,"prefix":"","firstName":"Nguyen","middleName":"Minh","lastName":"Son","suffix":""},{"id":594835779,"identity":"0f29f01e-3cd9-450a-9f35-e1975a48b5c8","order_by":3,"name":"Tran Tay","email":"","orcid":"","institution":"Lac Hong University","correspondingAuthor":false,"prefix":"","firstName":"Tran","middleName":"","lastName":"Tay","suffix":""},{"id":594835780,"identity":"0cedd23a-5c1f-4f45-8354-70bdafee8040","order_by":4,"name":"Thanh Q. Nguyen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxElEQVRIiWNgGAWjYHACNiA+IIdgEwbMYC3GDGzMJGpJbCBaizl7/rEHP3fcSd8u33+A4UPZYQaD2w34tVj2PGY37D3zLHdnGzMD44xzQC13DuDXYnAjmU2Ct+1w7oZjzAzMQAaD5IwEwlok/7YdTjcAaflLrBZpoOEJYC2MQC38EoS0nHlsJi3bdthww7Fkg4M959J5CGs5nvhM8m3bYXmDwwcfPvhRZi3HRkgLAwOSggNAzENIPaqWUTAKRsEoGAVYAQBNw0Myx4RK/QAAAABJRU5ErkJggg==","orcid":"","institution":"Trường ĐH Nguyễn Tất Thành","correspondingAuthor":true,"prefix":"","firstName":"Thanh","middleName":"Q.","lastName":"Nguyen","suffix":""}],"badges":[],"createdAt":"2025-12-21 01:08:16","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8414686/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8414686/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":103505862,"identity":"39cea92a-f233-4ebc-b9cc-010fbb4487a8","added_by":"auto","created_at":"2026-02-26 13:33:17","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2403097,"visible":true,"origin":"","legend":"","description":"","filename":"EnhancingImageR1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8414686/v1_covered_8914d826-4a64-465b-b5a1-b76ffee87f35.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Deep Siamese ResNet-50 Framework with Triplet loss for High-Precision Face Verification","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Image verification, Face recognition, Siamese network, ResNet-50, Triplet loss, Deep metric learning, Biometric authentication, Feature embedding, Few-shot learning, Model calibration","lastPublishedDoi":"10.21203/rs.3.rs-8414686/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8414686/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study proposes a deep learning framework for image verification that integrates a Siamese ResNet-50 architecture with triplet loss to enhance feature discrimination, particularly for facial recognition tasks. The model leverages the powerful residual connections of ResNet-50 for robust feature extraction and replaces the conventional contrastive loss with triplet loss to optimize inter-class and intra-class distances in the learned embedding space. Advanced training strategies, including the Adam optimizer, Cosine Annealing learning rate scheduling, and weight decay regularization, are employed to stabilize convergence and improve generalization. The proposed model is evaluated on the challenging Labelled Faces in the Wild (LFW) dataset, achieving 88.33% accuracy, an F1-score of 0.8828, and an AUC-ROC of 0.96. These results outperform baseline architectures such as Siamese VGG16, ResNet-34, and ConvNextTiny, while maintaining a favorable computational complexity. Additional analyses including ROC curves, mean Average Precision (mAP), inference time, and calibration performance demonstrate the model\u0026rsquo;s superior balance between accuracy, speed, and deployment readiness. These findings highlight the model's strong potential for practical biometric authentication systems and few-shot learning applications.\u003c/p\u003e","manuscriptTitle":"A Deep Siamese ResNet-50 Framework with Triplet loss for High-Precision Face Verification","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-24 10:34:49","doi":"10.21203/rs.3.rs-8414686/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"8524b298-713d-4970-a9c5-0b8d73dc9c2d","owner":[],"postedDate":"February 24th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-04-18T18:38:31+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-24 10:34:49","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8414686","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8414686","identity":"rs-8414686","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}