Research On Real-time Video Processing Algorithms

With the development of the multimedia technology, digital video systems are widely used in many applications, such as digital storage media, digital TV, and remote monitor & control system. The main object of this dissertation focuses on digital video. Video codec, video post-processing and video error concealment are studied.Firstly, video codec is studied. In order to meet customers' increasing demands of the bandwidth of network and the quality of video, different organizations have enacted next generation video standards such as H.264, AVS and VC-1. The video compression ratio is increased highly by these standards, but hardware implementation of video codec becomes much more difficult. In the dissertation, these new video standards are analyzed at the aspects of hardware cost, bus bandwidth, and the extent of the algorithms' complication. The method of reusing overlapped reference data is given to reduce the loading of bus.Existing context-adaptive variable length coding algorithms are inefficient, irregular and has high computation cost. In order to overcome these disadvantages, classified context-adaptive variable length coding algorithm is introduced in the dissertation. This algorithm codes level and run separately using inerratic code tables. All code words could be obtained through a few simple calculations avoiding the look-up table. The methods to code run are classified according to the known coding information, which improves the compression efficiency.Secondly, video post-processing is studied. In many applications, the formats of video should be converted in real-time, e.g. scaling and de-interlacing etc. The video post-processor, playing a role as an interface between video source and display device, is responsible for these processes. In order to reduce the computation cost of post-processing, decoder/post-processor co-architecture is proposed, in which the decoder should feed coding information to the post-processor. The coding information directs and optimizes post-processing. Based on this co-architecture, scaling algorithms based on coding information are presented. These algorithms analyze DCT coefficients and motion vectors, subsequently obtain texture characteristic of the picture. Then the algorithms select corresponding methods to process each area of the picture. Meanwhile, a de-interlacing algorithm based on coding information is also introduced. This algorithm can realize de-interlacing by motion compensation with low computation cost. In order to avoid using ill-suited coding information whenpost-processing, a series of methods are put forward to check coding information.Finally, video error concealment is studied. Compressed video data may be destroyed by various factors when they are stored or transferred. So error concealment is indispensable for an excellent video decoder. The recent domestic and international achievement is summarizes at first. Then, spatial error concealment is studied and a new algorithm based on edge extending is given. Strong edges across error macroblock are protected by the algorithm and picture details are recovered partly. Temporal error concealment is also studied, and new algorithms based on coding information are put forward. These algorithms modify motion vectors according to coding information, which increases the precision of motion vectors. Moreover, in order to reduce blocking distortion caused by temporal error concealment, a de-blocking method is present to smooth block edges and improve the appearance of the image.