LSSVM-based blind steganalysis for JPEG images
Pla Information · 2007
Least squares support vector machine(LSSVM)-based blind steganalysis for JPEG(joint photographic expert group) images is constructed.Each JPEG image is characterized using 18 dimensions calibrated features calculated from the DCT(discrete cosine transform) and the spatial domain.The LSSVM based classifier trained with feature vectors corresponding to cover and stego images,whose output set by different embedding methods,can be applied to multi-class blind steganalysis.Binary classifier and quaternion classifier steganalysis against the popular steganography algorithms such as Jsteg,F5 and MB is performed.The experimental results show that it is effective to classify the cover and stego images using the binary classifier,and the multi-classifier is reliable in classifying high embedding rate stego images.