Scalable Video Coding Research Based On Video Content

As an extension of H.264/AVC, SVC (Scalable Video Coding) has been researched and applied widely due to its high-quality video service for different networks and terminal equipments. Highly coding efficient and user video quality can not be achieved without the consideration of video content. In this paper, we make a deep research on SVC with focuses on video content. An Eq-MSE (Equivalent Mean Square Error) method and the MinFR (Minimal Frame Rate Without Jitter) parameter have been proposed. For lower computing complex, a simplified rate model have been proposed. For transmission rate and coding time, the effect of GOPsize has been researched, and find an optimized GOPsize. Using the proposed method and model, the performance and quality of the SVC will be increased effeciently. The main contributions and specific work of this thesis are shown as following:(1) An Eq-MSE method is proposed for calculating the spatial frequency of non-sine picture. The ratio between the MSE value of adjacent frames within video sequence and unit space frequency at the same condition’s is the spatial frequency of2D non-sinusoidal signal. The spatial frequency is the basis of computing the temporal frequency of video sequence. The relationship between spatial frequency and perceptual quality is analyzed in detail, then draw a conclusion:when direction of the object movement is identical to the spatial frequency, video sequence has the maximum temporal frequency. While they are orthogonal, the temporal frequency equal to0. That means no matter how fast the object moves, human will not perceive its movement.(2) The parameter MinFR is defined, which is the minimal frame rate could be accepted by human visual. And its calculation method is proposed. If the frame rate is lower than MinFR, the movement of objects in video sequence looks discontinuous. While if the frame rate is larger than MinFR, the movement looks continuous without jitter. MinFR is obtained through the temporal frequency of video sequence. In order to reduce complexity of computing, the simplified calculation method is proposed. With the simplified method, a satisfied frame rate can be achieved accurately. It’s important in practice for proposing MinFR, which makes the extraction of sub-stream with optimal quality become possible, under a restricted bandwidth.(3) A simplified rate model with temporal and SNR scalability is presented. The map between temporal frequency and temporal scalability parameter of model is constructed, then the parameter will be got by a looking-up table. It reduces the complexity of the rate model. We verify MinFR on subjective evaluation, and the experimental results demonstrate that the sub-stream of the combination of temporal and SNR enhancement layers can get the optimal perceptual quality with the MinFR under a given bandwidth.(4) The rate and the time models about GOPsize are proposed, which can exactly depict the relationship between rate, time and GOPsize, respectively. A combined model with rate and time is presented to choose GOPsize adaptively to achieve the optimal tradeoff between coded stream rate and coding time. Based on the combined model, the method of adjusting parameter is proposed to adapt different system requirements. Meanwhile, the validity of the joint model is tested at the special condition that the parameters of stream rate and coding time are the same.