Adversarial Image Integrity and Tamper Check Browser Extension

Abstract

Abstract - Client-side filtering tools typically protect user privacy through local transformations of media files. However, existing tools depend on third-party trust or operate outside of the daily upload workflow. We developed a Chrome Extension that leverages Five-Layer Adversarial Pixel Perturbation and casts transformations on user images directly in the browser, preventing the original unprotected image from being sent to a server. This protection is unobtrusive and introduces structures that degradation can be manifested through significant manipulation of the media; including cropping, resizing, recompressing, or AI-enhancements. We implemented four different methods for browser-level upload interception, utilizing dynamic input detection through a MutationObserver. We confirmed no visible perturbation and degradation was manifested through user interaction with the Extension for tamper detection verification.