Steganalysis for JPEG Images Using Extreme Learning Machine

This paper proposes a novel blind Steganalysis process, for colored JPEG images. Extreme Learning Machine (ELM) has been used in the paper to classify the images into stego images and non-stego images. The feature set used for classification of images consists of 810 features. First 405 features are based on Markov random process applied on correlations among JPEG coefficients of image. Calibration is applied on these Markov features to get the remaining 405 features. These calibrated features are the difference between the Markov features of the image and Markov features of a reference image, obtained by decompressing, cropping and recompressing the image. Experimental results show that our proposed ELM based steganalysis method clearly outperforms other SVM based steganalysis methods in terms of percentage of correctly classified images and in terms of time taken for both training and testing. The fast speed of the proposed method due to fast learning time of ELM makes it useful for real-time steganalysis.