Light field image compression with sub-apertures reordering and adaptive reconstruction

Abstract

Light field (LF) attracts tremendous attention due to its capability of recording the intensity of scene objects as well as the direction of the light ray, which also dramatically increases the amount of redundant data. In this paper, we explore the structure of the light field images, and propose a pseudo-sequence based light field image compression with sub-aperture reordering and adaptive reconstruction to efficiently improve the coding performances. In the proposed method, we firstly decompose the lenslet image into sub-aperture images, and then design an optimized sub-aperture scan order to rearrange them sequentially as a pseudo-sequence. Third, we take advantage of the state-of-the-art video codec to compress the pseudo-sequence by leveraging both intra- and inter-view correlations. Considering the interpolation and transform induced by the reconstruction procedure from sub-aperture images to lenslet image, we propose an enhanced reconstruction method by applying region-based non-local adaptive filters which extracts the non-local similarities for collaborative filtering to promote the quality of reconstructed lenslet images. Extensive experimental results show that the proposed method achieves up to 15.7% coding gain in terms of BD-rate.