Efficient Depth Image Compression Using Accurate Depth Discontinuity Detection and Prediction

This paper presents a novel depth image compression algorithm for both 3D Television (3DTV) and Free Viewpoint Television (FVTV) services. The proposed scheme adopts the K-means clustering algorithm to segment the depth image into K segments. The resulting segmented image is losslessly compressed and transmitted to the decoder. The depth image is then compressed using a bi-modal block encoder, where the smooth blocks are predicted using direct spatial prediction. On the other hand, blocks containing depth discontinuities are approximated using a novel depth discontinuity predictor. The residual information is then compressed using a lossy compression strategy and transmitted to the receiver. Simulation results indicate that the proposed scheme outperforms the state of the art spatial video coding systems available today such as JPEG and H.264/AVC Intra. Moreover, the proposed scheme manages to outperform specialized depth image compression algorithms such as the one proposed by Zanuttigh and Cortelazzo.