Steganalysis of LSB Matching Based on the Sum Features of Average Co-occurrence Matrix Using Image Estimation

Abstract

A new LSB matching steganalysis scheme for gray images is proposed in this paper. This method excavates the relevance between pixels in the LSB matching stego image from the co-occurrence matrix. This method can acquire high accuracy near to 100% at high embedding rate. In order to increase the accuracy at low embedding rate, we strengthen the differences between the cover image and the stego image to improve the performance of our scheme. Two 8 dimensional feature vectors are extracted separately from the test image and the restoration image, and then the combining 16 dimensional feature vector is used for steganalysis with the FISHER linear classification. Experimental results show that the detection accuracy of this method is above 90% with the embedding rate of 25%; even when the embedding rate is 10%, the detection accuracy reaches 80%.Experiments show that this method is more reliable than other state-of-art methods.