为了更好地平衡隐写容量和不可感知性之间的关系,本文提出了一种基于SteganoGAN的优化方案.首先,将SteganoGAN隐写网络和提取网络进行加深,以增强模型的复杂度和学习能力;其次,为了实现更为隐蔽的信息嵌入,在隐写网络部分引入离散小波变换(Discrete Wavelet Transformation,DWT)和逆离散小波变换(Inverse Discrete Wavelet Transform,IDWT)模块,这使得秘密信息能够被有效地嵌入到图像的小波域中;最后,在隐写网络、提取网络中融入了一种改进的通道-空间注意力模块(Improved Channel and Spatial Attention Module,ICAM-SAM),促使模型能够聚焦于图像中的高隐蔽性区域,实现更为精准的信息隐藏.实验结果表明:改进后的模型在提取准确率上提高了0.84百分点,表明其隐写和提取过程更加精确.此外,每像素嵌入率(Reed-Solomon Bits-Per-Pixel,RS-BPP)提高了1.71%,这表明改进后的模型在相同大小的图像中可以隐藏更多的信息.同时,峰值信噪比(Peak Signal to Noise Ratio,PSNR)提高了12.53%、结构相似性(Structural Similarity Index,SSIM)提高了5.14%,这表明嵌入的信息对原始图像的影响更小,改进后的模型具有更高的图像质量.综合结果表明,改进后的模型具有更好的不可感知性和较大的隐写容量.
Image Steganography Method Based on Attention Mechanism and Wavelet Transform
In order to better balance the relationship between steganographic capacity and imperceptibility,this paper proposed an optimization scheme based on SteganoGAN.Firstly,the steganography network and extraction network of SteganoGAN were deepen to enhance the complexity and learning capability of the model;Secondly,in order to achieve more covert information embedding,discrete wavelet transform(DWT)and inverse discrete wavelet transform(IDWT)modules were introduced in the steganography network,which enabled secret information to be effectively embedded in the wavelet domain of the image;Finally,an improved channel and spatial attention module(ICAM-SAM)was incorporated into the steganography and extraction networks,enabling the model to focus on highly covert areas in the image and achieve more accurate information hiding.Experimental results show that the accuracy of the improved model is increased by 0.84 percentage point,indicating that its steganography and extraction pro-cesses are more accurate.In addition,the reed-solomon bits per pixel(RS-BPP)is increased by 1.71%,indicating that the improved model can hide more information in images of the same size.Meanwhile,the peak signal to noise ratio(PSNR)is increased by 12.53%and the structural similarity index(SSIM)is increased by 5.14%,indicating that the embedded information has a smaller impact on the original image and the improved model has higher image quality.In summary,these results indicate that the improved model has better imperceptibility and larger steganography capacity.
万方数据
维普
