Steganalysis for LSB Matching by Counting Image Blocks with the Same Gray Level

The steganalysis for the least significant bit (LSB) matching is a hot and difficult topic. Some bit plane/s might he changed during the LSB matching steganograpy. Huang et al. proposed a method based on the differences in the least two bit planes of an image and the results show good performance to uncompressed images. In this paper, we make some improvements by introducing Median Filter and sum image. The Median Filter, which is effective to remove random noise in an image, is useful for decreasing the noise brought by embedding based on the additive noise steganography model. The sum image is the same as a low-pass filter to an image in some degree. With Median Filter and sum image together, the influence to a stego image is more than a cover image so the differences can be used for discriminating. Extensive experimental results show that the new detectors in our improved method perform better than the old ones, especially to uncompressed images.