{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/neural-imaging-pipelines-the-scourge-or-hope","title":"Neural Imaging Pipelines - the Scourge or Hope of Forensics?","arxiv_id":"1902.10707","date":"2019-02-27","proceeding":null,"authors":["Pawel Korus","Nasir Memon"],"abstract":"Forensic analysis of digital photographs relies on intrinsic statistical\ntraces introduced at the time of their acquisition or subsequent editing. Such\ntraces are often removed by post-processing (e.g., down-sampling and\nre-compression applied upon distribution in the Web) which inhibits reliable\nprovenance analysis. Increasing adoption of computational methods within\ndigital cameras further complicates the process and renders explicit\nmathematical modeling infeasible. While this trend challenges forensic analysis\neven in near-acquisition conditions, it also creates new opportunities. This\npaper explores end-to-end optimization of the entire image acquisition and\ndistribution workflow to facilitate reliable forensic analysis at the end of\nthe distribution channel, where state-of-the-art forensic techniques fail. We\ndemonstrate that a neural network can be trained to replace the entire photo\ndevelopment pipeline, and jointly optimized for high-fidelity photo rendering\nand reliable provenance analysis. Such optimized neural imaging pipeline\nallowed us to increase image manipulation detection accuracy from approx. 45%\nto over 90%. The network learns to introduce carefully crafted artifacts, akin\nto digital watermarks, which facilitate subsequent manipulation detection.\nAnalysis of performance trade-offs indicates that most of the gains can be\nobtained with only minor distortion. The findings encourage further research\ntowards building more reliable imaging pipelines with explicit\nprovenance-guaranteeing properties.","url_abs":"http://arxiv.org/abs/1902.10707v1","url_pdf":"http://arxiv.org/pdf/1902.10707v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"neural-imaging-pipelines-the-scourge-or-hope","repo_url":"https://github.com/pkorus/neural-imaging","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"image-manipulation","task_name":"Image Manipulation"},{"task_slug":"image-manipulation-detection","task_name":"Image Manipulation Detection"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}