A new copy move forgery detection technique with automatic threshold determination

Ensuring authenticity of images has become an important issue recently. Copy move forgery is one of the most common tampering techniques used to modify images. Copy move forgery detection techniques in the literature divide the image into overlapping blocks and use various techniques to extract features from the blocks. Similarity between the feature vectors is a clue about the forgery. However, these techniques use a predefined threshold to test the similarity. Test images with different characteristics require various threshold values. Determination of the best threshold value can be troublesome because the range of the feature vector elements’ cannot be predetermined. Therefore, many experiments must be realized to find the best threshold value. In this work, we utilize DCT-phase terms to restrict the range of the feature vector elements’ and Benford’s generalized law to determine the compression history of the image under test. The method uses element-by-element equality between the feature vectors instead of Euclidean distance or cross correlation and utilizes compression history to determine the threshold value for the current test image automatically. Experimental results show that the method can detect the copied and pasted regions under different scenarios and gives higher accuracy ratios/lower false negative compared to similar works.