GABIC: Graph-based Attention Block for Image Compression

TL;DR

GABIC introduces a graph-based attention mechanism for neural image compression, reducing feature redundancy and improving performance, especially at high bit rates, by integrating k-Nearest Neighbors into attention modules.

Contribution

The paper proposes GABIC, a novel graph-based attention block that effectively reduces feature redundancy in neural image compression models.

Findings

GABIC outperforms comparable methods at high bit rates.

The method enhances compression efficiency by reducing redundant features.

Experimental results demonstrate improved performance over existing approaches.

Abstract

While standardized codecs like JPEG and HEVC-intra represent the industry standard in image compression, neural Learned Image Compression (LIC) codecs represent a promising alternative. In detail, integrating attention mechanisms from Vision Transformers into LIC models has shown improved compression efficiency. However, extra efficiency often comes at the cost of aggregating redundant features. This work proposes a Graph-based Attention Block for Image Compression (GABIC), a method to reduce feature redundancy based on a k-Nearest Neighbors enhanced attention mechanism. Our experiments show that GABIC outperforms comparable methods, particularly at high bit rates, enhancing compression performance.