Zhang Xi, Wu Xiaolin
IEEE Trans Image Process. 2021;30:963-975. doi: 10.1109/TIP.2020.3040074. Epub 2020 Dec 8.
We propose a novel asymmetric image compression system of light l -constrained predictive encoding and heavy-duty CNN-based soft decoding. The system achieves superior rate-distortion performances over the best of existing image compression methods, including BPG, WebP, FLIF and recent CNN codecs, in both l and l error metrics, for bit rates near or above the threshold of perceptually transparent reconstruction. These remarkable coding gains are made by deep learning for compression artifact removal. A restoration CNN is designed to map a lossy compressed image to its original. Its unique strength is to enforce a tight error bound on a per pixel basis. As such, no small distinctive structures of the original image can be dropped or distorted, even if they are statistical outliers that are otherwise sacrificed by mainstream CNN restoration methods.
我们提出了一种新型的非对称图像压缩系统，该系统采用轻量级 l 约束预测编码和基于重型卷积神经网络（CNN）的软解码。在 l 和 l 误差度量标准下，对于接近或高于感知透明重建阈值的比特率，该系统在现有最佳图像压缩方法（包括 BPG、WebP、FLIF 和近期的 CNN 编解码器）之上实现了卓越的率失真性能。这些显著的编码增益是通过深度学习来去除压缩伪像实现的。设计了一个恢复 CNN，将有损压缩图像映射回其原始图像。其独特优势在于在每个像素基础上强制实施严格的误差界限。因此，即使原始图像中的小的独特结构是统计异常值，否则会被主流 CNN 恢复方法舍弃或扭曲，它们也不会被丢弃或扭曲。
