Non-linear Bi-directional prediction for depth coding

Abstract

A depth image represents a relative distance from a camera to an object in the three-dimensional (3-D) space and it is widely used as 3-D information in computer vision and computer graphics. Generally, the depth is represented as an image format and it is uniformly quantized in the disparity/intensity domain whereas it is non-uniformly quantized in the depth domain. Thus, the conventional bi-prediction applied in the disparity/intensity domain does not catch up the value for the linearly moving object. To solve this problem, we propose a non-linear bi-directional prediction for depth coding. Experimental results demonstrate that the proposed non-linear bi-directional prediction method achieves by 0.68 dB of the PSNR gain over the conventional method when the hierarchical-B picture coding is used. © 2009 Springer-Verlag Berlin Heidelberg.