基于深度SR模型的加密数字图像压缩与重构

Compression and Reconstruction ofEncrypted Digital Image based on Deep SR Model

针对图像压缩后降低存储空间的同时也降低图像分辨率的问题,提出一种基于深度超分辨率(SR)模型的加密数字图像压缩与重构方法。先对加密数字图像进行分割,然后针对分割后的图像子块进行编码压缩处理,并将经典的SR重构方法(稀疏编码法)与深度学习(卷积神经网络)进行结合,构建一种深度SR模型,并利用模型对图像进行压缩和解压,最后对解密后的数字图像进行重构。结果表明:图像压缩后较压缩前占据存储空间降低,压缩效果有所改善,经过深度SR模型重构后的数字图像分辨率相对更高,且峰值信噪比更高。

In order to solve the problem of reducing the storage space and image resolution after image compression,a method of compression and reconstruction of encrypted digital image based on deep super resolution(SR)model was proposed.Firstly,the encrypted digital image was segmented,and the image sub-blocks after segmentation were encoded and compressed.Secondly,the classical SR reconstruction method(sparse coding method)was combined with deep learning(convolutional neural network)to construct a deep SR model,and this model was used to compress and decompress the image.Finally,the digital image after decryption was reconstructed.The results show that the image occupies less storage space after compression than before compression,and the compression effect is improved.The resolution of digital image after deep SR model reconstruction is relatively higher,and the peak signal-to-noise ratio is higher.