Behaviour knowledge space-based fusion for image forgery detection

The detection of manipulated images is of utmost significance nowadays, mainly due to the various technological, moral, and judicial connotations associated with image editing. In this paper, a common form of image manipulation known as image composition or splicing is examined. The forgery detection method proposed here makes use of fine inconsistencies in the colour of the lighting of images. Our approach is machine-learning based which makes use of a behaviour knowledge space. This technique can be used in photographs having two or more people. To achieve forgery detection, complementary illuminant estimators are incorporated on image regions of similar material. Colour-, texture- and edge-based features are extracted from these illuminant estimates which are then provided to a machine-learning approach for automatic decision-making.