Compression of depth information for 3D rendering

Abstract

This paper presents a novel strategy for the compression of depth maps. The proposed scheme starts with a segmentation step which identifies and extracts edges and main objects, then it introduces an efficient compression strategy for the segmented regions' shape. In the subsequent step a novel algorithm is used to predict the surface shape from the segmented regions and a set of regularly spaced samples. Finally the few prediction residuals are efficiently compressed using standard image compression techniques. Experimental results show that the proposed scheme not only offers a significant gain over JPEG2000 on various types of depth maps but also produces depth maps without edge artifacts particularly suited to 3D warping and free viewpoint video applications.