An Ensemble Model using CNNs on Different Domains for ALASKA2 Image Steganalysis

Abstract

We present our third place solution for the ALASKA2 Image Steganalysis competition. We develop detectors using convolutional neural networks (CNNs) on both the spatial domain and the frequency domain of the discrete cosine transform used in JPEG compression. Our CNN detectors use state-of-the-art architectures in image classification tasks. We adjust the architecture to better capture the features of steganography methods in the frequency domain. We build an ensemble model of these CNNs, in which both spatial and frequency domain models contribute to performance. In this paper, we describe those models in detail and explain how the techniques used in them improve accuracy through experiments.