Research On Optimization Of Spatial And Temporal Prediction, In-loop Filtering For H.264/AVC Codec

Since the digitized video information involves data of huge volumes, video compression is indispensable for the convenience of storing and transmitting the digital video signals. In the viewpoint of information theory, the essence of video compression is to reduce the redundancy that exists in the video images. A series of video coding standards established by the International Standardization Organization (ISO/IEC) and International Telecommunication Union (ITU-T) make the application of video compression algorithms greatly successful. To improve the compression efficiency further, the latest video compression standard - H.264, which is recommended by ISO/IEC and ITU-T jointly, adopts several original techniques to squeeze the redundancy.The intra prediction algorithm in spatial domain, where lots of prediction modes are introduced, is adopted by H.264 to improve compression ratio of intra-frame coding by reducing the spatial pixels-correlative redundancy. However, the augment of computation complexity and storage space resulted from the multimode decision constrains the implementation of video codec in the embedded system. Utilizing the directivity of prediction mode to reduce the candidates, a novel fast algorithm for intra prediction, in which the real-time codec implementation and storage space requirement are considered, is proposed in chapter 2. This fast algorithm can reduce the computation complexity and storage space to a great degree while maintaining the image quality and compression ratio.Motion vectors with quarter-sample resolution for the luminance component in H.264 makes the offset between matched blocks and current blocks much smaller, thus the coding bit-rate is reduced efficiently. Nevertheless, the computation complexity to create the non-existing pixels by interpolation increases many-fold. Based on the instruction set of the embedded multimedia processor - MediaDSP32, interpolation algorithm optimization is given in the chapter3. Meanwhile, to improve the codec performance by reducing the delay time in data interchange, an adaptive pixel / macro-block level boundary extension strategy is also proposed, which lowers the frequency of data transaction between on-chip and off-chip memory, and enhances the memory bandwidth utilization to a great extent.Block-based hybrid coding is the mainstream technology of video coding now and in the near future. Since the blocks are compressed relatively separated, there is a severe defect in such technology, namely obvious blocking artifacts will appear when the compression ratio is high, thus the subjective quality of video images will be debased. In-loop De-blocking filter adopted by H.264 can solve such defect efficiently, Howerver;a great deal of data transaction is needed in the filtering operation, which restrains the improvement of codec performance. To lower the frequency of data scheduling, a more efficient memory allocation strategy, in which the data correlation of intra prediction and in-loop de-blocking filtering is considered deliberately, is proposed in the chapter 4.