抄録
This paper describes a method for detecting a signature from BPCS-Steganography by attacking it with a statistical analysis of images. BPCS is a technique that hides a large amount of data in digital images. It divides bit-planes produced through bit-plane decomposition from the images into sub-binary images, and embeds noise-like secret data in the sub-images. These sub-images are then extracted by comparing measures known as complexity and threshold values. We consider a complexity histogram representing the relative occurence frequency of the various noisy regions in each bit-plane. The complexity histograms of the sub-images, in which secret data has been embedded, nearly fit a normal distribution for the image containing secret data. Using the correlation coefficients between the complexity histograms, regarded as a BPCS signature, this method distinguishes between natural images and images with secret data embedded. In our experiments, we were able to accurately detecte these signatures using this method.
