ProxIQA: A Proxy Approach to Perceptual Optimization of Learned Image Compression

TL;DR

This paper introduces ProxIQA, a proxy network that aligns image compression optimization with human perceptual quality, leading to significant bitrate reductions while maintaining perceptual quality.

Contribution

ProxIQA provides a novel proxy-based perceptual optimization framework for learned image compression, improving bitrate efficiency over traditional MSE-based methods.

Findings

Achieved up to 31% bitrate reduction at the same perceptual quality.

Demonstrated effectiveness of ProxIQA in end-to-end image compression training.

Validated the approach using VMAF as a perceptual quality metric.

Abstract

The use of $\ell_p$ $(p=1,2)$ norms has largely dominated the measurement of loss in neural networks due to their simplicity and analytical properties. However, when used to assess the loss of visual information, these simple norms are not very consistent with human perception. Here, we describe a different "proximal" approach to optimize image analysis networks against quantitative perceptual models. Specifically, we construct a proxy network, broadly termed ProxIQA, which mimics the perceptual model while serving as a loss layer of the network. We experimentally demonstrate how this optimization framework can be applied to train an end-to-end optimized image compression network. By building on top of an existing deep image compression model, we are able to demonstrate a bitrate reduction of as much as $31\%$ over MSE optimization, given a specified perceptual quality (VMAF) level.