摘要：

伪装图像质量、恢复图像质量和传输安全性是图像隐藏最关注的3个问题。为解决这些问题, 提出一种基于深度可逆网络和差分编码的图像隐藏方法, 并用于保护电力巡检缺陷图像。首先训练深度可逆网络, 利用训练好的可逆缩放网络对电力巡检缺陷图像进行向下缩放。与压缩感知等方法相比, 可逆缩放网络能够恢复质量更高的缺陷图像。然后提出一种新的基于差分编码的嵌入算法, 利用该算法将下缩放的缺陷图像嵌入到封面图像中。不同于现存很多方法直接对原图像像素值进行嵌入, 所提方法先利用差分编码对缺陷图像进行编码, 然后利用最低有效位算法完成嵌入操作, 差分编码后的图像数值集中在更小的范围内, 减少了嵌入对封面图像像素值的损害。实验结果表明, 相较对比方法, 所提方法伪装图像的峰值信噪比(PSNR)提高3.99 dB~16.56 dB, 恢复缺陷图像的PSNR提高12.52 dB~17.02 dB。另外, 该方法对SPAM的抗隐写分析性能优于对比方法。分析结果表明, 所提方法在伪装图像质量、恢复缺陷图像质量和传输安全性方面的表现优于许多先进方法。

关键词: 电力系统, 电力缺陷图像, 可逆缩放网络, 差分编码, 隐写分析

Abstract:

Stego image quality, recovered image quality, and transmission security are the three most important factors in image hiding. This study proposes an image-hiding method based on a deep invertible network and differential coding to protect defective images from power inspection. First, a deep invertible network is trained. A trained invertible rescaling network downscales defective images. An invertible scaling network recovers defective images with higher quality compared to methods such as compressive sensing. Next, a new embedding algorithm based on differential coding embeds the downscaled defective image into the cover image. Unlike existing methods that directly embed the original image pixel values, this study first encodes the defective image using differential coding and then performs the embedding using the least significant bit algorithm. After differential coding, the image values are concentrated in a smaller range, reducing the damage caused by embedding pixel values into the cover image. In experiments, the Peak Signal-to-Noise Ratio (PSNR) of the stego image improved by 3.99 dB-16.56 dB, and the PSNR of the recovered defective image improves by 12.52 dB-17.02 dB. This method outperforms the comparative method in terms of anti-steganalysis performance for SPAM. Experimental results show that the proposed method improves the quality of the stego image, the recovered defective image, and transmission security, outperforming many state-of-the-art methods.

Key words: power systems, power defect image, invertible rescaling network, differential encoding, steganalysis