Block-based Illumination Compensation and Search Techniques for Multiview Video Coding

We consider block-based interview coding techniques for multiview video sequences that are robust to illumination variations across views. We consider both global illumination differences which could be due to lack of camera calibration or heterogeneity, as well as local ilumination changes caused by lack of camera aligment. We propose a two-parameter model for illumination compensation (IC) that is used at the block matching search step and can be adaptively applied by taking into account the rate-distortion characteristics of each block. We also present a modified search (MS) method for block-based interview disparity estimation that makes use of feature points in each frame in order to provide a good prediction of likely disparity vectors. We present coding experiments on different multiview sequences based on the H.264/AVC reference codec. The proposed techniques show significant coding gains for interview coded frames, as compared to methods that do not employ IC and MS.