New blind steganalysis for breaking F5 algorithm
Tao Zhang · Jisuanji gongcheng yu sheji · 2009
Generally, steganalysis based on statistical characteristic has a strong pertinency, and a blind steganalysis can adapt itself to the steganography. Hereby, a steganalysis for breaking F5 on the basis of these merits is given. The 21 features are extracte for detecting. SVM is used to classify, which weakly depends on the quantity and quanlity of training samples. The experimental results that are done on the different conditions show that the algorithm woks well. Especially for detecting images with embedding rate as low as 25%, and the detecting rate keeps more than 95% when keeping the false alarm rate less than 4%.