Steganalysis of Colored JPEG Images

Images, especially colored JPEG images, are increasingly being used as cover by many steganography techniques. Steganalysis detects the presence of message embedded by steganography techniques. Many authors have given different steganalysis techniques which differ mainly on the feature sets being used. In this paper we are comparing feature sets of six of these steganalysis techniques for multiclass JPEG steganalysis. The work considers three JPEG steganography techniques (i.e. F5, OutGuess and Model-Based) and different embedding rates for creating stego images. SVM, a binary classifier by nature, is used as classifier and to handle multiple classification task of multi-class steganalysis One-Against-All technique is applied. Experimental results show that the ability of steganalysis schemes to detect stego-images vary for different steganography techniques. Also, the performance of steganalysis schemes show upward trend with increase in the embedding rate.