Saliency Driven Perceptual Image Compression

TL;DR

This paper introduces a novel end-to-end trainable image compression model that emphasizes perceptual similarity and saliency, outperforming traditional metrics and enhancing visual quality and computer vision task performance.

Contribution

It presents a new perceptual similarity metric, integrates saliency into compression, and demonstrates improved visual quality and downstream task performance.

Findings

Outperforms existing methods in visual quality.

Provides better results for object detection and segmentation.

Proposes a learned perceptual similarity metric.

Abstract

This paper proposes a new end-to-end trainable model for lossy image compression, which includes several novel components. The method incorporates 1) an adequate perceptual similarity metric; 2) saliency in the images; 3) a hierarchical auto-regressive model. This paper demonstrates that the popularly used evaluations metrics such as MS-SSIM and PSNR are inadequate for judging the performance of image compression techniques as they do not align with the human perception of similarity. Alternatively, a new metric is proposed, which is learned on perceptual similarity data specific to image compression. The proposed compression model incorporates the salient regions and optimizes on the proposed perceptual similarity metric. The model not only generates images which are visually better but also gives superior performance for subsequent computer vision tasks such as object detection and segmentation when compared to existing engineered or learned compression techniques.