A Performance Study of ±1 Steganalysis Employing a Realistic Operating Scenario

We present results of an experiment to detect the presence of ±1 embedding in digital images employing a realistic operating scenario. The ±1 embedding steganography (also known as LSB matching) is simple to implement but not easy to detect. Many of the results presented in the literature to date demonstrate detection performance for a system that has been trained on a particular type of image, camera and/or embedding rate. In real world operations these parameters may not be known. In this paper, we begin by comparing our previously developed steganalysis method to others in the literature using an identical image data set and assuming a known embedding rate. We show that the performance of our technique compares favorably to the state of the art. Subsequently, we demonstrate the impact on our steganography detector when the embedding rate is not known or mismatched. Our experiments show the range of performance degradation that is incurred from training the detector with a mismatched embedding rate. We then build a global detector that is trained using images with several steganographic embedding rates. Results show a small decrease in performance when employing the global detector. In most cases the performance of the global detector performs better than other embedding-rate-mismatched detectors for the suspect images.