Saliency-aware End-to-end Learned Variable-Bitrate 360-degree Image Compression

TL;DR

This paper introduces a novel saliency-aware end-to-end learned compression model for 360-degree images that allocates bits based on region importance, significantly reducing data size while preserving visual quality.

Contribution

It is the first to propose an end-to-end variable-rate model for 360-degree image compression that incorporates saliency information.

Findings

Achieves significant bit-rate savings compared to state-of-the-art methods.

Maintains high perceptual visual quality with reduced data size.

Addresses the unique properties of omnidirectional images for efficient compression.

Abstract

Effective compression of 360$^\circ$ images, also referred to as omnidirectional images (ODIs), is of high interest for various virtual reality (VR) and related applications. 2D image compression methods ignore the equator-biased nature of ODIs and fail to address oversampling near the poles, leading to inefficient compression when applied to ODI. We present a new learned saliency-aware 360$^\circ$ image compression architecture that prioritizes bit allocation to more significant regions, considering the unique properties of ODIs. By assigning fewer bits to less important regions, significant data size reduction can be achieved while maintaining high visual quality in the significant regions. To the best of our knowledge, this is the first study that proposes an end-to-end variable-rate model to compress 360$^\circ$ images leveraging saliency information. The results show significant bit-rate savings over the state-of-the-art learned and traditional ODI compression methods at similar perceptual visual quality.