WaSP: Hierarchical Warping, Merging, and Sparse Prediction for Light Field Image Compression

Abstract

We propose a versatile light field compression scheme that is organized on hierarchical levels, where all views belonging to a particular level are encoded using several views already encoded in the previous hierarchical levels. The new scheme builds on an earlier version of our codec, and provides a more generalized functionality with improved view merging. The operations needed when one view is encoded conditional on its reference views are: first warping its reference views to the location of the current view and partitioning the pixels according to their state of occlusion in various warped versions; then merging the warped references using one optimal LS merger for each class of occluded pixels; finally, adjustment of the overall merged image to the original view by using a sparse predictor. The new scheme is applied to both plenoptic camera images and high density camera array data, and is evaluated in accordance with the JPEG Pleno test conditions. We compare the performance of the proposed codec to that of the HEVC anchors defined in the JPEG Pleno test conditions. We also make comparisons to the performance achieved by our earlier scheme. The proposed codec is publicly available on GitHub and it was accepted as the Verification Model (VM) 1.0 software for JPEG Pleno Light Field coding standard.