Research On Key Techniques Of Object-based Video Scalable Coding In The Wavelet Domain

Fine granular scalable video coding is a difficulty and research hotspot in the state-of-art of video coding discipline. And there are few researches on the object-based video coding in the literature. So the research on the object-based fine granular scalable video coding is not only important in theory, but also in urgent needs. This paper makes a deep insight into two problems of object-based fine granular scalable video coding in the wavelet domain, namely video object coding and motion estimation techniques. As to video object coding, this paper first proposes a shape adaptive integer wavelet transform (IWT) algorithm based on lifting scheme, and analyses the distribution characteristics of the transform coefficients, then a quantization threshold scheme based on odd number square and a quantization strategy based on binary search are proposed. On the basis of above, a shape adaptive EZW based on IWT and the new binary quantization scheme is proposed. Simulation results prove that the proposed scheme can achieve 0. 5~2dB higher PSNR than traditional threshold and bit-plane based quantization scheme at the same bit-rate. In terms of motion estimation techniques, this paper takes the development of the motion estimation techniques in the wavelet domain as the thread to sum up the present techniques as three kinds, namely, band to band motion estimation techniques, indirect band motion estimation techniques and shift-variant resistant motion estimation techniques, and then makes a wide and comparative study on the three kinds of methods. Based on the deep study above, an MLBS (Modified-Low-Band-Shift) motion estimation scheme is proposed. Then this paper puts forward an MLBS-based scalable video motion estimation scheme in the wavelet domain (MLBSSME). The MLBS method improves the accuracy of the motion estimation. And the "blocking artifacts" at the finest resolution are effectively reduced by overlapped block motion compensation technique. According to the multi-resolution representation characteristics of wavelet transform, the estimation process is limited in a search window of reduced size so that the search speed is accelerated. Experimental results show the motion compensation quality of MLBSSME is nearly the same as LBS, but the time and space complexity of MLBSSME is only 30%-40% as much as the complexity of LBS. The proposed scheme can be used widely in the spatial scalable, rate (SNR) scalable and hybrid spatial and rate scalable video codec.