关键词: 图像隐写分析, 混合滤波器, 通道注意力机制, 深层卷积, 深度可分离卷积

Abstract: Aiming at the problems of large number of parameters, limited generalization ability and low accuracy of current image steganalysis model, a lightweight image steganalysis model based on multidirection hybrid filters is constructed. In this model, a multidirectional and multisize high and low frequency hybrid filter bank is designed and the channel attention mechanism is used to preprocess the image, so as to adaptively extract the effective features in the image and improve the generalization ability of the model. The feature extraction module designs a multilayer convolution including the residual module to extract the image features in depth and improve the ability of the model to capture features. The dimensionality reduction module adopts depthwise separable convolution instead of traditional convolution, which effectively reduces the number of model parameters and achieves lightweight. Comparative analysis of experimental data showed that the model had the characteristics of lightweight and good generalization ability, and improved the accuracy of steganalysis.

Key words: image steganalysis, hybrid filters, channel attention mechanism, deep convolution, depthwise separable convolution