Payload Length and Location Identification Using a Novel CNN and Re-embedding Strategy in Stego Images Created by Content Adaptive Steganographic Algorithms

With the rapid development of technology in the modern digital world, covert communication of secret information as a payload without instigating visible attention by using steganography emerged as a possible threat. Steganalysis is the counter attack to steganography. Research is being carried out vastly to differentiate an innocent cover image from stego image and identifying steganographic method while very few attempts have been made to identify payload length and location information. In this work, a novel convolutional neural network regressor is proposed to identify the embedding change rate which gives the length of payload. In addition to identifying the suspicious content length, location of the payload bits is identified using a re-embedding algorithm from the flipped locations of stego images. Then this information is used to provide an estimate of the hidden secret. Experiments are carried out using spatial content-adaptive algorithms namely Highly Undetectable SteGO with Bounding Distortion (HUGO-BD), Wavelet Obtained Weights (WOW) and Spatial Universal Wavelet Distortion (S-UNIWARD) on BOSSBase database with five payload bins such as 0.1, 0.2, 0.3, 0.4 and 0.5 bpp. From the experimental results it is shown that the proposed model outperforms state-of-the-art steganalysis features and regressors.