SADN: Learned Light Field Image Compression with Spatial-Angular Decorrelation

TL;DR

This paper introduces SADN, a novel neural network for light field image compression that effectively decouples spatial and angular information, achieving significantly higher compression efficiency and quality compared to existing standards and methods.

Contribution

SADN is the first end-to-end network to jointly decorrelate spatial and angular information for light field image compression, outperforming traditional codecs and existing neural networks.

Findings

Achieves 2.137x and 2.849x higher compression efficiency than H.266/VVC and H.265/HEVC.

Provides 79.6% bitrate savings over end-to-end image compression networks.

Outperforms existing methods in subjective quality and light field consistency.

Abstract

Light field image becomes one of the most promising media types for immersive video applications. In this paper, we propose a novel end-to-end spatial-angular-decorrelated network (SADN) for high-efficiency light field image compression. Different from the existing methods that exploit either spatial or angular consistency in the light field image, SADN decouples the angular and spatial information by dilation convolution and stride convolution in spatial-angular interaction, and performs feature fusion to compress spatial and angular information jointly. To train a stable and robust algorithm, a large-scale dataset consisting of 7549 light field images is proposed and built. The proposed method provides 2.137 times and 2.849 times higher compression efficiency relative to H.266/VVC and H.265/HEVC inter coding, respectively. It also outperforms the end-to-end image compression networks by an average of 79.6% bitrate saving with much higher subjective quality and light field consistency.