Research On Algorithm And Application For New Generation Video Compression Standards

With the development of the multimedia and internet technology, digital video compression is widely used in many applications, such as video phone, video conference, IPTV and digital TV. In order to meet customers' increasing demands on video applications, many standardization organizations have enacted many new video standards. The standards of H.264, AVS and VC-1, which are called the next generation video standards because of the more excellent performance compared with the traditional ones, are widely studied nowadays.Taking the new generation video standards as reach object, the author analyzed and studied the issues of video codec framework, inter-frame prediction, intra-frame prediction, transform and deblocking filter, and proposed several optimum algorithms and realization structures, most of which have been verified on the FPGA or ASIC with steady performance.Firstly, inter-frame prediction and intra-frame prediction for H.264 are studied. An adaptive threshold early termination search (ATETS) algorithm based on quantization step is proposed for integer-pixel motion estimation, and a mode correlation based fractional-pixel search (MCFPS) algorithm is proposed for fractional-pixel motion estimation. For intra-frame prediction, a novel policy based on mode statistic and transform coefficient all-zero block (MSZB) is presented. With these optimum algorithms, experimental results show that the complexity of inter-frame and intra-frame prediction is reduced notably with negligible degradation in image quality and compression rate. The optimum H.264 encoder can perform video compression as twice fast as the original one.Secondly, the video codec framework is studied. In order to meet the characteristic of the new generation video standard, many different video codec frameworks have been proposed. In this thesis, an implemented framework based on MPU and ASIC for multi-standard video decoding is proposed. Besides the advantage of the ASIC design, a novel mechanism of hardware-software codesign is applied within the framework, which makes it more competitive in cost and performance.Thirdly, transform and inverse transform are studied. The new generation video standards apply several different kinds of transform, such as 8x8DCT, 8x8IntegerDCT, 4x4IntegerDCT, 8x4IntegerDCT and 4x8IntegerDCT. A normalized 8x8 transform algorithm is proposed by reconstructing the 8x8 pixel blocks and transform matrixes, which is applicable to all standards. Then, an implemented structure is presented, which can meet the requirement of transform process while reducing the hardware resources efficiently.Fourthly, the multi-standard video deblocking filter is studied. By reconstructing the data structure, a novel filtering order is proposed for both H.264 and AVS. For VC-1, optimum filtering orders for overlap-filtering and deblock-filtering are presented by extending the reference data. With the proposed filtering order, the filtering operation can be performed on each macroblock instead of the whole picture. Then, an implemented structure is introduced, which can efficiently reduce the operations, data throughput and hardware resources. In addition, a deblocking filter algorithm for error concealment is presented to improve the appearance of the image.It is the summary and expectation at the end of the thesis.