This article has presented a new steganalysis method based on wavelet transform. First, wavelet transform is applied on the original image to separate different frequencies subbands. Then, for greater focus on frequency regions, the blocking technique is carried out on these subbands and they are divided into equal sized sectors. Afterwards, optimal wavelet packet transform method is applied in feature extraction stage. Entropy cost function of coefficients is modified with message embedding and tends to be zero. As a result, the ultimate optimal tree having the maximum number of necessary nodes can be built. Experimental results indicate that the ultimate obtained features are a good basis to detect a secret message in the image.

No abstract

Feature extraction is the base of steganalysis which is a part of image processing research field. This article has proposed a steganalysis method for digital images. Common steganalysis techniques go over the entire image; this will reduce their focus on higher frequencies in which there is a higher probability for hidden messages. Accordingly, in this article, images are first decomposed into smaller blocks and then optimal wavelet packet decomposition method is applied to extract the features of each block. In the proposed algorithm, characteristic function moments obtained from wavelet sub-bands are used as features. These features are arranged in a tree structure and then an entropy cost function is used to select the optimal values of these features. In the next step, the blocks are classified in several categories and a classifier appropriate to the features of each category is applied to distinguish cover or stego blocks. Finally, the majority vote rule is applied on the results obtained from the blocks to determine whether the entire image is a cover or stego image. The experimental results of this steganalysis method show its high accuracy as compared to the common steganalysis algorithms in the frequency domain.