Empirical Rate-Distortion Analysis of JPEG 2000 3D and H. 264/AVC Coded Integral Imaging Based 3D-Images

Novel camera systems producing 3D-images containing light direction in addition to light intensity is emerging. Integral imaging (II) is a technique on which many of these systems rely. The pictures produced by these cameras (II -pictures) are space-requiring in terms of data storage compared to their 2D counterparts. This paper investigates how coding the II-pictures using H.264/AVC and JPEG 2000 Part 10 (JP3D) affect the images in terms of rate-distortion as well as introduced coding artifacts. A set of four reference images are coded using a number of pre-processing and encoding variants, so called coding schemes. For low bitrates (<0.5 bpp) the H.264/AVC-based coding schemes have higher coding efficiency, which asymptotically level of at higher bitrates in favor of JP3D. The JP3D coded 3D-images show less spread in quality than H.264/AVC, when quality is evaluated using PSNR as a function of viewing angle. However, the distortion induced by H.264/AVC is primarily localized to object boarders within the 3D-image, which in initial tests appear less visible than the JP3D coding artifacts that spread out evenly over the image. Extensive subjective tests will be performed in future work to further support the presented results.