Benchmarking steganographic and steganalysis techniques

There have been a number of steganography embedding techniques proposed over the past few years. In turn the development of these techniques have led to an increased interest in steganalysis techniques. More specifically Universal steganalysis techniques have become more attractive since they work independently of the embedding technique. In this work, our goal is to compare a number of universal steganalysis techniques proposed in the literature which include techniques based on binary similarity measures, wavelet coefficients' statistics, and DCT based image features. These universal steganalysis techniques are tested against a number of well know embedding techniques, including Outguess, F5, Model based, and perturbed quantization. Our experiments are done using a large dataset of JPEG images, obtained by randomly crawling a set of publicly available websites. The image dataset is categorized with respect to the size and quality. We benchmark embedding rate versus detectability performances of several widely used embedding as well as universal steganalysis techniques. Furthermore, we provide a framework for benchmarking future techniques.