Research On Vector Quantization Applications In Video Compression

Digital video processing is the core of multimedia information and the one of most challenging research fields. The large amount of digital videos is a huge challenge in the digital video development. And to solve this, instead of only relying on the development of network bandwidth and hardware technology, data compression is considered as an effective technique. Video compression helps us to make use of the transmission channel and storage resource more efficiently. Although the storage and the capacity of transmission have been enlarging, video coding will still be one of essential problems of multimedia research and applications in the foreseeable future.As an efficient technology for lossy data compression, vector quantization (VQ) is widely used in the field of image coding with virtues of simple principle and high compression rate. VQ is based on the Shannon's rate-distortion theory. VQ finds the nearest codeword for each input vector and transmits the corresponding index to the decoder, thus in the decoding phase only a simple table-look-up operation is needed. With the development of VQ techniques, more and more scholars pay attention to VQ based video compression algorithms. Compared with traditional algorithms, some of them have a clear theory background with high coding performance. This thesis investigates the theory, concepts, key techniques of VQ, and analyzes the theory and design methods of VQ based video compression. The thesis proposes a novel video compression scheme based on predictive vector quantization. Experiment results show it is effective.The research work in this thesis includes:This thesis summarizes the development and state-of-art of VQ technology and video coding techniques systematically; especially laying emphasis on VQ based video compression. We aim to find out the development of VQ algorithms and video compression techniques,for the purpose of combining VQ and video coding algorithms.This thesis introduces the main international video codec standards, especially the ITU-H.264 prediction and motion compensation algorithm. The framework of H.264 is different from previous standards, but it improves function units so that H.264 has the highest coding efficiency, yet with highest complexity. Generally speaking, H.264 is still a video compression standard based on transformation domain.Aiming at circumventing the weakness of high complexity of H.264 and the advantage of simple structure, low coding complexity and high rate-distortion performance of VQ technology, this thesis proposes a novel predictive VQ based video compression scheme, which uses H.264 standards coding method for reference, combining VQ techniques. It uses the minimax partial distortion competitive learning method to generate a public codebook, making sure that only codeword indices need to be transmitted after video sequence compression. It uses the mean-distance-ordered partial codebook search algorithm for fast code vector searching, which uses the mean of input vectors to reduce the computing expense dramatically without performance loss.Experimental results show that this scheme, with low computation complexity and high coding efficiency, makes a good tradeoff between rate and distortion. It can be used in real time applications and mobile handsets with limited computing capability.