A Steganalysis Algorithm Based on Denoising of Source Image Using ICA and SVM

In this paper we propose a new method for feature extraction in the context of still image steganalysis. At first, a denoising algorithm is employed to generate a new version of the observation form the original image. FastICA is employed to separate two sources from the two versions of the input image. Features are extracted from these two estimated sources. At the end support vector machine (SVM) is used as classifiers. This supervised learning method classifies the input image into either stego-image or clean-image. The performance of this algorithm is verified using some test samples. The results of our empirical tests show that detection accuracy of our method reaches 72% for true positive and 79% for true negative when 100% capacity of image is used for steganography.