Analysis and Comparison of Modern Video Compression Standards for Random-access Light-field Compression
Permanent address of the item is
Light-field (LF) 3D displays are anticipated to be the next-generation 3D displays by providing smooth motion parallax, wide field of view (FOV), and higher depth range than the current autostereoscopic displays. The projection-based multi-view LF 3D displays bring the desired new functionalities through a set of projection engines creating light sources for the continuous light field to be created. Such displays require a high number of perspective views as an input to fully exploit the visualization capabilities and viewing angle provided by the LF technology. Delivering, processing and de/compressing this amount of views pose big technical challenges. However, when processing light fields in a distributed system, access patterns in ray space are quite regular, some processing nodes do not need all views, moreover the necessary views are used only partially. This trait could be exploited by partial decoding of pictures to help providing less complex and thus real-time operation. However, none of the recent video coding standards (e.g., Advanced Video Coding (AVC)/H.264 and High Efficiency Video Coding (HEVC)/H.265 standards) provides partial decoding of video pictures. Such feature can be achieved by partitioning video pictures into partitions that can be processed independently at the cost of lowering the compression efficiency. Examples of such partitioning features introduced by the modern video coding standards include slices and tiles, which enable random access into the video bitstreams with a specific granularity. In addition, some extra requirements have to be imposed on the standard partitioning tools in order to be applicable in the context of partial decoding. This leads to partitions called self-contained which refers to isolated or independently decodable regions in the video pictures. This work studies the problem of creating self-contained partitions in the conventional AVC/H.264 and HEVC/H.265 standards, and HEVC 3D extensions including multi-view (i.e., MV-HEVC) and 3D (i.e., 3D-HEVC) extensions using slices and tiles, respectively. The requirements that need to be fulfilled in order to build self-contained partitions are described, and an encoder-side solution is proposed. Further, the work examines how slicing/tiling can be used to facilitate random access into the video bitstreams, how the number of slices/tiles affects the compression ratio considering different prediction structures, and how much effect partial decoding has on decoding time. Overall, the experimental results indicate that the finer the partitioning is, the higher the compression loss occurs. The usage of self-contained partitions makes the decoding operation very efficient and less complex.