Joint Learned and Traditional Image Compression for Transparent Coding

Abstract

This paper proposes a novel image compression framework, which consists of a CNN-based method and a versatile video coding (VVC) based method. The CNN-based method uses the auto-encoder to learn the quantized latent representation of the image and joints the autoregressive and hierarchical priors to exploit the probabilistic structure. We also design a post-processing network for VVC to further improve the quality of compressed images. We find that CNN-based method and VVC-based method are complementary to each other in terms of MS-SSIM and PSNR. Thus, we combine the two methods together to obtained better coding performance. Furthermore, to select the best compression parameter, an optimal coding mode selection algorithm is introduced. Experimental results indicate that the proposed image compression scheme can achieve significantly better rate-distortion (RD) performance than other methods.