Scalable video coding using spatiotemporal interpolation

This thesis addresses the problem of scalable video coding for multimedia visual communication applications. Multimedia applications often share the same video signals at different resolutions. Although a simple solution to multi-resolution video coding is the simulcast technique, in which each resolution of video is coded independently, a more efficient alternative is scalable video coding. In scalable video coding, the bandwidth allocated to a lower-resolution reproduction of the video signal is partially re-utilized in coding higher-resolution video. Error resilience and complexity division are the two important features of scalable coding schemes.; Currently, scalability features are only included in the MPEG-2 video coding standard. In MPEG-2, scalability was explored in five different regimes: spatial, temporal, SNR, frequency, and data partitioning. This thesis investigates scalable coding within the context of MPEG-2. Among these five scalability types, spatial and temporal scalability are analyzed in detail, while brief descriptions of the others are also included.; The concept of inter-layer prediction plays the central role in spatially and temporally scalable coding schemes. Considering a basic scalability structure with two layers, inter-layer prediction is employed to reduce the redundancy in the enhancement layer. In this thesis, a novel approach to nonlinear interpolation is developed by introducing a novel spatio-temporal interpolation technique and a deinterlacing algorithm derived from it. The spatial and temporal predictions are combined in a spatially varying fashion to yield an effective prediction scheme. Then, these novel nonlinear video interpolation techniques are employed as the inter-layer prediction units in various spatially and temporally scalable video coding applications. Scalable coding schemes are analyzed for both progressive and interlaced format picture processing. Within the context of scalable coding, various issues related to inter-laced coding are identified and presented. For the application of the novel nonlinear interpolation methods, the MPEG-2 scalability context is chosen due to its inherent scalability features and provisions for interlaced processing, however they can be equally applied within the other video coding standards, such as the MPEG-4 content-based coding standard, at a variety of bit rates.; The prediction ability of the the nonlinear spatio-temporal video interpolation techniques in scalable coding were demonstrated by comparing them with the simulcast technique (temporal prediction) both visually and based on PSNR results. An average of 2 to 3 dB higher PSNR is achieved by scalable coding compared to the simulcast technique.