3D Geometry Compression with Hybrid Framework: Quasi-JPEG and Phase Encoding

Wonbeen Oh, Jae-Sang Hyun

Abstract: This paper proposes a novel end-to-end compression framework and implementation of depth image using phase encoding. Our method divides depth image into stair map and periodic map using sinusoidal phase encoding, and utilize lossless compression for stair map and lossy compression for periodic map to reduce errors on depth discontinuities. Our algorithm uses Run Length Encoding and entropy encoding for lossless compression of stair map. Periodic map is subdivided using trigonometric functions, and Discrete Cosine Transform was used for lossy compression. Experimental results show our method can achieve higher PSNR quality than JPEG2000, when compression was done on same bit-rate.