Abstract: The existing local feature point-based methods such as scale-invariant feature transform (SIFT) and speeded up robust features (SURF) cannot be accurately extracted as feature points especially in terms of forgeries involving small areas and smooth regions. A two-stage forgery detection method which is based on the hue, saturation and value (HSV) color space and contrast limited adaptive histogram equalization (CLAHE) was proposed. In the first stage, the tested image is converted from red, green and blue (RGB) color space to HSV and then SURF features were extracted in this space. In the second stage, in order to resist tampering in smooth regions, CLAHE algorithm was used as a preprocessing stage to the SURF algorithm. After feature extraction, generalized 2 nearest neighbor (g2NN) matching skills were used in the matching step. At last, the random sample consensus (RANSAC) was used to remove the false matches. The location of the tampered area is achieved by morphological operations. Experiments show that the proposed algorithm not only can resist copy-move forgeries in small areas and flat regions with non-significant visual features but also robust to post-processing operations such as rotation, scaling and so on.

Key words: hue, saturation and value color space, contrast limited adaptive histogram equalization, two-stage feature detection