Image coding algorithms for video-conferencing applications
Ming-Hong Chan · Spiral (Imperial College London) · 1989
This thesis is concerned with low rate video coding, a special branch of image coding with emphasis on real-time, low bit-rate performance.Intended applications include video-conferencing and video-telephony, areas that involve video signals with relatively low motion content.The basic approach adopted in this work involves the segmentation of an image into two classes of regions i) those that can be motion compensated and ii) those that cannot be motion compensated.For regions which can be motion compensated, predicted signal values are generated from the previous frame and used to approximate the current frame.For regions which cannot be motion compensated intra frame coding techniques are used.Attempts by earlier workers involved the use of efficient high compression techniques for still images to code the difference signals for video sequences, but have shown little success.Indeed the use of these techniques produce impairments that are visually disturbing, and oftentimes unacceptable, and which could not be directly anticipated from single frame image coding.In this work the nature and cause of such distortion are identified.Moreover, several ways of achieving improvements are proposed.These include better modelling of the frame difference signal, the use of motion compensation, high level approaches and spatio-temporal coding techniques.Through analysis and experimentation it becomes evident that for applications, such as video-conferencing, which involve a high degree of temporal redundancy, extensive' motion compensation can be used to advantage.An improved motion compensation algorithm is developed which overcomes the shortcomings of conventional motion compensation schemes.The algorithm attempts to segment an image into regions of uniform motion by adaptively dividing an image into rectangular blocks of variable size.From the studies carried out for this thesis, it is apparent that while motion compensation is effective for the major part of an image, those image parts for which motion compensation fails should be coded in an intra-frame manner.This leads to a revised scheme involving an intra/inter-frame switch within the segmentation process.Intra-frame coding is incorporated in a coherent manner, resulting in a single-pass codec capable of producing pictures of good visual quality while operating at very low rates.It is thus established that intra/inter-frame (intensity/motion) segmentation is a feasible way towards low bit-rate, good quality video coding.