Random Forest based Copy-Move Forgery Detection
Proving the authenticity of images is animportant part of image forensics. Copy-move forgery is amethod of forgery commonly followed in blind image forensics.We propose the use of a modified Auto Color Correlogram toobtain feature vectors from the forged image. The featuresextracted are sent as input to a RBF-SVM that gives a score forthe possibility of a copy-move situation. We then use anormalized cross correlation for feature matching with thesame feature vectors and then produce texture attributes assmoothness and Entropy. Based on the entropy andsmoothness we use a linear regression model to classify thisand obtain a predicted score. The two outputs obtained arepassed as input to a Random forest classifier which classifiesthe image as either forged or not forged.