Abstract
In this paper, we tend to analyze one amongst the foremost common styles of photographic manipulation,called image composition or splice. We tend to propose a forgery detection methodology that exploits refinedinconsistencies within the color of the illumination of pictures. Our approach is machine-learning primarily basedand needs borderline user interaction. The technique is applicable to pictures containing 2 or a lot of folks and needsno professional interaction for the meddling call. To attain this, we tend to incorporate info from physicsandstatistical-based fuel estimators on image regions of comparable material. From these fuel estimates, we tend toextract texture- and edge-based options that square measure then provided to a machine-learning approach forautomatic decision-making. The classification performance victimization associate degree SVM meta-fusion classifieris promising.