Deep Neural Network Watermarking Based on Texture Analysis

Abstract

In recent years, deep neural network is active in the field of computer image vision. The existing digital watermarking technology based on deep neural network can resist image attacks, but image quality is not satisfied. In order to improve the quality of the watermarked images, a neural network watermarking method based on image texture analysis is proposed. Firstly, image texture features are analyzed by gray co-occurrence matrix, and the image is divided into texture complex region and flat region. Secondly, in order to reduce the degree of image modification for better quality, StegaStamp network is adopted to embed the watermark into the flat texture area. Finally, from the perspective of traditional multiplicative watermarking embedding, the watermark embedding process of deep neural network is improved to enhance the watermarked image quality. Experimental results show that the proposed method can effectively improve the quality of the watermarked images without degrading the robustness.