Adversarial Sample Generation Method for Scene Text Images Based on Up-sampling

Adversarial Sample Generation Method for Scene Text Images Based on Up-sampling | 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 Adversarial Sample Generation Method for Scene Text Images Based on Up-sampling Zihao Zeng, Chenxiao Wang, Xiao Yang, Yong Chen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5287761/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 Scene Text Recognition (STR) has significantly enhanced the efficiency of information acquisition and interaction in natural environments. However, it also introduces potential security risks, such as the unauthorized recognition and extraction of sensitive textual information from the environment, including personal identification numbers, license plates, and other confidential data. To address privacy protection in scene text, recent research has proposed using minimal pixel perturbations to safeguard textual information, making the content observable but difficult to extract accurately. However, such perturbation attacks are easily noticeable to the human eye, allowing adversaries to counteract them with defensive measures such as filtering small perturbations. Existing methods fail to simultaneously ensure high visual quality and make the perturbation attacks imperceptible. In this study, we propose a novel scene text adversarial sample generation method incorporating up-sampling. This method achieves a high attack success rate while increasing the payload applied to the image, preserving the perturbed image quality, and improving the stealthiness of the adversarial samples. To further enhance the quality of the perturbed images, we introduce the Adaptive Local Search Attack (ALSA), which utilizes adaptive perturbation based on visual quality and perceptual loss to ensure that the perturbed image remains as similar as possible to the original image in human vision, which can further enhance the stealthiness of adversarial samples and make perturbation attacks difficult to detect. Our experimental results show that the proposed method maintains high visual quality while achieving a better protection success rate across various text recognition models compared to existing methods. up-sampling scene text recognition black-box adversarial examples 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-5287761","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":368275770,"identity":"0504f832-83b5-46d5-bc8e-c1f35acffc2a","order_by":0,"name":"Zihao Zeng","email":"","orcid":"","institution":"Chongqing Normal University","correspondingAuthor":false,"prefix":"","firstName":"Zihao","middleName":"","lastName":"Zeng","suffix":""},{"id":368275772,"identity":"bf1ef09a-ba3a-4016-9546-ef70bf7bea36","order_by":1,"name":"Chenxiao Wang","email":"","orcid":"","institution":"Chongqing Normal University","correspondingAuthor":false,"prefix":"","firstName":"Chenxiao","middleName":"","lastName":"Wang","suffix":""},{"id":368275774,"identity":"cac512a7-e556-491e-9c6f-c529f0b649d5","order_by":2,"name":"Xiao Yang","email":"","orcid":"","institution":"Chongqing Normal University","correspondingAuthor":false,"prefix":"","firstName":"Xiao","middleName":"","lastName":"Yang","suffix":""},{"id":368275775,"identity":"76eefdab-b741-48b4-8ec4-c977e114ed5e","order_by":3,"name":"Yong Chen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6ElEQVRIie3NsWrDMBCAYQmBsgiynmnyDgcCZyntq0gU/Aith0BVDMqWrgn0ITxlVhHEix4go0IgUx+gW1ucDh2K7DGDfm467uMIyeWusCmjTVS4fH79XVAzRIpVYzHWe7o1YwmGzhQxMNq6sYQctEFtOZOd350EuZ23jp1jStCNNlHbGS9D9SQFqWTr+AJThMHliygPorwRxOvWCQ4pwn8IaMtAbnryNUyEeDegAkOEnrhhApMXi6reKwjVY/GGD3LreZkk935yOn7iUk1Xfgcf9d183TXnJPkb9kPY2PsLyeVyudw/fQPDGEihaQkKBwAAAABJRU5ErkJggg==","orcid":"","institution":"Chongqing Normal University","correspondingAuthor":true,"prefix":"","firstName":"Yong","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2024-10-18 08:23:28","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5287761/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5287761/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":68877041,"identity":"4137b547-3bf6-43f4-bd52-94d8175723b4","added_by":"auto","created_at":"2024-11-13 04:54:05","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":554962,"visible":true,"origin":"","legend":"","description":"","filename":"AdversarialSampleGenerationMethodforSceneTextImagesBasedonUpsampling.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5287761/v1_covered_189c567b-0eec-4e5b-a006-468676fd94c5.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Adversarial Sample Generation Method for Scene Text Images Based on Up-sampling","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":"up-sampling, scene text recognition, black-box, adversarial examples","lastPublishedDoi":"10.21203/rs.3.rs-5287761/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5287761/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eScene Text Recognition (STR) has significantly enhanced the efficiency of information acquisition and interaction in natural environments. However, it also introduces potential security risks, such as the unauthorized recognition and extraction of sensitive textual information from the environment, including personal identification numbers, license plates, and other confidential data. To address privacy protection in scene text, recent research has proposed using minimal pixel perturbations to safeguard textual information, making the content observable but difficult to extract accurately. However, such perturbation attacks are easily noticeable to the human eye, allowing adversaries to counteract them with defensive measures such as filtering small perturbations. Existing methods fail to simultaneously ensure high visual quality and make the perturbation attacks imperceptible. In this study, we propose a novel scene text adversarial sample generation method incorporating up-sampling. This method achieves a high attack success rate while increasing the payload applied to the image, preserving the perturbed image quality, and improving the stealthiness of the adversarial samples. To further enhance the quality of the perturbed images, we introduce the Adaptive Local Search Attack (ALSA), which utilizes adaptive perturbation based on visual quality and perceptual loss to ensure that the perturbed image remains as similar as possible to the original image in human vision, which can further enhance the stealthiness of adversarial samples and make perturbation attacks difficult to detect. Our experimental results show that the proposed method maintains high visual quality while achieving a better protection success rate across various text recognition models compared to existing methods.\u003c/p\u003e","manuscriptTitle":"Adversarial Sample Generation Method for Scene Text Images Based on Up-sampling","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-10-30 04:16:54","doi":"10.21203/rs.3.rs-5287761/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":"5e55a366-6b66-4fcf-bac5-0ea92b3173df","owner":[],"postedDate":"October 30th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-11-13T04:53:48+00:00","versionOfRecord":[],"versionCreatedAt":"2024-10-30 04:16:54","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5287761","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5287761","identity":"rs-5287761","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}