Non-local Attention Optimized Deep Image Compression

TL;DR

This paper introduces a deep image compression framework that uses non-local attention mechanisms within a variational auto-encoder to better capture correlations and adapt bit allocation, leading to improved compression performance.

Contribution

It presents a novel NLAIC framework integrating non-local operations and attention in VAE-based image compression, enhancing entropy coding and feature importance estimation.

Findings

Outperforms existing learned and conventional methods on Kodak dataset.

Achieves higher PSNR and MS-SSIM metrics.

Effectively captures local and global correlations for better compression.

Abstract

This paper proposes a novel Non-Local Attention Optimized Deep Image Compression (NLAIC) framework, which is built on top of the popular variational auto-encoder (VAE) structure. Our NLAIC framework embeds non-local operations in the encoders and decoders for both image and latent feature probability information (known as hyperprior) to capture both local and global correlations, and apply attention mechanism to generate masks that are used to weigh the features for the image and hyperprior, which implicitly adapt bit allocation for different features based on their importance. Furthermore, both hyperpriors and spatial-channel neighbors of the latent features are used to improve entropy coding. The proposed model outperforms the existing methods on Kodak dataset, including learned (e.g., Balle2019, Balle2018) and conventional (e.g., BPG, JPEG2000, JPEG) image compression methods, for both PSNR and MS-SSIM distortion metrics.