Light field compression using homography and 2D warping

This paper describes a new method for light field compression that exploits inter-view correlation. The proposed method uses homography and 2D warping to predict views and does not require additional camera parameters or a 3D geometry model. The method utilizes angular shift between views, which is neglected in conventional motion compensation methods. Results indicate improved coding efficiency of the proposed method over traditional motion compensation schemes. A full light field coder based on video-compression demonstrates 1-1.5 dB additional improvement in PSNR, or equivalently a 4-12% additional reduction in bitrate when the new method is introduced as a prediction mode.