Image Tampering Localization Based on Superpixel Segmentation

With the rapid development of image editing software such as Photoshop and GIMP, image tampering becomes simpler than ever before without living obvious traces. Photo response non-uniformity (PRNU) is a kind of camera fingerprint, which has always been an effective tool for image forgery detection. By detecting the correlation between camera PRNU and image PRNU, we can judge whether there is tampering. Unlike existing PRNU-based tampering localization algorithms where it estimates correlation based on a sliding window, this paper uses the simple linear iterative clustering (SLIC) superpixel algorithm for non-overlapped segmentation of the detected image. The correlation of PRNU is calculated on the superpixel to determine whether there is tampering. The difference between superpixel and sliding window is that superpixel can accurately depict the edge of the tampered objects, especially for the added object type of tampering. Extensive experimental results demonstrate that the proposed algorithm provides much better localization performance than the state-of-art sliding window based methods with less complexity.