基于密集残差网络的图像隐藏方案

针对基于编-解码器网络的图像隐写方案生成的含密图像和消息图像质量不高的问题,提出了一种新的基于密集残差连接的编码器-解码器隐写方案,与现有的端到端图像隐写网络不同,本文采用密集残差连接,将浅层网络的特征输送到深层网络结构的每一层,有效的保留了特征图的细节信息,并使用通道和空间注意力模块对特征进行筛选,提高了编-解码器对图像复杂纹理区域的关注度。在LFW、PASCAL-VOC12和ImageNet数据集的实验结果表明,在保证算法安全性的前提下,所提方法能够有效提高图像质量,含密图像和载体图像的峰值信噪比(PSNR)和结构相似性(SSIM)的平均值最高达到了36.2dB和0.98。

Image Hiding Scheme Based on Dense Residual Network

A new encoder decoder steganography scheme based on dense residual connections is proposed to address the problem of low quality of encrypted and message images generated by image steganography schemes based on encoder decoder networks. Unlike existing end-to-end image steganography networks, this paper uses dense residual connections to transport the features of shallow networks to each layer of deep network structures, effectively preserving the detailed information of feature maps, And the channel and spatial attention modules are used to filter features, improving the attention of the encoder decoder to complex texture areas of the image. The experimental results on LFW, PASCAL-VOC12, and ImageNet datasets show that the proposed method can effectively improve image quality while ensuring algorithm security. The average peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of the encrypted and carrier images can reach up to 36.2dB and 0.98, respectively.