Convolutional Neural Network with Multi-scale Pooling for the Efficient Steganalysis in Images of Arbitrary Sizes

Numerous research studies have consistently demonstrated that convolutional neural networks (CNNs) outperform traditional machine learning methods that employ a two-part structure for detecting the presence of data hidden under a steganography approach known as steganalysis. Existing CNN models for the steganalysis of digital images use several approaches such as data augmentation, absolute value function and others to enhance the classification accuracies. Nevertheless, many state-of-the-art approaches rely on stacking numerous convolutional layers to expand the local receptive fields. However, these approaches showed a common drawback to not effectively extracting features from stego images. In this article, we propose a CNN with multi-scale pooling for efficient steganalysis in images of arbitrary sizes. We take advantage of small convolutions and extensively explore convolution's salient features such as Xception and Inception, and apply the spatial pyramid pooling. Through experimentation, the results show the outperformance of the proposed method compared to two existing methods. The results also highlight the proposed CNN's effectiveness and versatility in steganalysis images with arbitrary size.