Exposing Photo Manipulation Using Geometry and Shadows

摘要

It is increasingly easier to manipulate digital images by the sophisticated photo editing software. Often visual inspection cannot definitively distinguish manipulation from authentic images. This paper introduces a forensic technique that focuses on geometric and shadow color inconsistencies which arise when fake objects with shadows are inserted into an image or an object with its shadow in the image are modified. This paper analyzes three underlying geometric relations and shadow color characteristic constrains that occur in image scene. In particular, (ⅰ ) explore the property of vanishing point in linear perspective project, and evaluate the geometric consistent level of the image based on the uncertain degree of vanishing point;(ⅱ ) analyze the relation between illuminated object and its cast shadow which are modeled by the planar homology and use this constrain to estimate the image's geometric consistent level;(ⅲ ) locate tempered region by measure the K-L divergence between shadow pairs. Visually plausible forgery images demonstrate the performance of our proposed method.