Detecting Digital Image Forgeries using Color Constancy

UAbstract. There is a gradual increase in the number of composite pictures containing people. Due to the existence of such compositions, the trust in photographs is reduced. Due to the invention of powerful digital image editing tools, it has been so easy to manipulate images. Approaches that consider the illumination inconsistencies in digital images are of particular interest because a perfect illumination adjustment in a digital composite is very difficult to obtain. The proposed method is build upon the ideas by local illuminant color estimation of scenes, edge-based and texture-based color constancy analysis. Here, an illumination information provided by statistics based color constancy method is used. The method requires minimal user interaction for tampering decisions. For this, first the illuminant color is estimated using a statistical gray edge method, and treat this illuminant map as texture maps, then extract information using HOGedge algorithm and gabor texture features. These informations are then provided to a machine learning approach for automatic decision making. The classification is based on the well-known SVM classifier