Synonymous Variational Inference for Perceptual Image Compression

TL;DR

This paper introduces a novel variational inference framework based on semantic synonymity for perceptual image compression, providing theoretical insights and a new compression scheme that balances rate, distortion, and perception effectively.

Contribution

It proposes a synonymous variational inference approach that re-analyzes perceptual image compression through semantic synonymity, leading to a new compression scheme and theoretical understanding.

Findings

Theoretical proof of the perception compression tradeoff.

Development of a progressive SIC codec.

Comparable rate-distortion-perception performance achieved.

Abstract

Recent contributions of semantic information theory reveal the set-element relationship between semantic and syntactic information, represented as synonymous relationships. In this paper, we propose a synonymous variational inference (SVI) method based on this synonymity viewpoint to re-analyze the perceptual image compression problem. It takes perceptual similarity as a typical synonymous criterion to build an ideal synonymous set (Synset), and approximate the posterior of its latent synonymous representation with a parametric density by minimizing a partial semantic KL divergence. This analysis theoretically proves that the optimization direction of perception image compression follows a triple tradeoff that can cover the existing rate-distortion-perception schemes. Additionally, we introduce synonymous image compression (SIC), a new image compression scheme that corresponds to the analytical process of SVI, and implement a progressive SIC codec to fully leverage the model's capabilities. Experimental results demonstrate comparable rate-distortion-perception performance using a single progressive SIC codec, thus verifying the effectiveness of our proposed analysis method.