Wavelet-Based Multiresolution Features for Detecting Duplications in Images

Duplication of image regions is a common method for manipulating original images using typical software like Adobe Photoshop. In this study, we propose a wavelet based feature representation scheme for detecting duplicated regions in images. This technique works by first applying multi-resolution wavelet decomposition to small fixed-sized image blocks. Normalized wavelet coefficients are then stacked into a vector in an order from lower to higher frequencies. This kind of representation appears robust to block matching. Duplicated regions are then detected by lexicographically sorting all of the image blocks and applying threshold to the desired frequency of the offsets of the blockcoordinates. A semi-automatic technique that detects accurate number of duplicated regions is also proposed. Initial experiments with a set of natural images having duplicated regions show impressive results compared to linear PCA based representation.