Refining Coded Image in Human Vision Layer Using CNN-Based Post-Processing

TL;DR

This work addresses the gap of applying post-processing to scalable image coding for both humans and machines. It introduces a CNN-based post-processing module built on Residual in Residual Dense Blocks (RRDB) into a two-LIC scalable coding pipeline to learn and mitigate the distinctive coding noise seen in human-decoded outputs. Experiments on the COCO dataset show PSNR gains when using post-processing, with performance improving as the RRDB count increases, demonstrating improved human-viewable quality without compromising machine-relevant information. The approach offers a practical path to dual-purpose image coding by combining scalable compression with targeted post-processing to enhance perceptual quality.

Abstract

Scalable image coding for both humans and machines is a technique that has gained a lot of attention recently. This technology enables the hierarchical decoding of images for human vision and image recognition models. It is a highly effective method when images need to serve both purposes. However, no research has yet incorporated the post-processing commonly used in popular image compression schemes into scalable image coding method for humans and machines. In this paper, we propose a method to enhance the quality of decoded images for humans by integrating post-processing into scalable coding scheme. Experimental results show that the post-processing improves compression performance. Furthermore, the effectiveness of the proposed method is validated through comparisons with traditional methods.

Figures (3)

• Figure 1: Image compression process and structure of the post-processing model.
• Figure 2: Compression performance of the proposed and comparative methods. "sic” stands for scalable image coding.
• Figure 3: An example of the decoded images with and without post-processing and the coding noise in those images. Areas with large coding noise are shown brighter.