Research On Predictive Encoding Techniques For Wireless Video

By signal prediction of spatial domain and motion prediction compensation of time domain, predictive encoding technique realizes the elimination of the spatial and temporal information redundancy, and thereby reduces the volume of data needed by video signals presentation. Predictive encoding technique is a key component of the digital video compression, and has a profound impact on the compression efficiency and video quality. With wireless video communication's flourishing development, video compression technology, especially predictive encoding technology is facing a new challenge. First, to reduce the demand for expensive wireless network bandwidth, more advanced predictive encoding techniques with high computational complexity are used in video coding to increase the efficiency of compression. But the lack of computing power for wireless devices to achieve real-time predictive coding algorithm is a huge challenge. Second, in wireless computing environment, the uncertainty of computing ability puts a higher demand for the adaptive capacity of video encoders and predictive coding techniques face challenges ability to ensure the achievement and maintenance of real-time video quality in conditions of variable computing power. Therefore, fast and efficient predictive encoding techniques with the ability to adapt computing environment is research focus and study of coding technique for wireless video.Based on the current popular predictive encoding structure, this paper focused on the study of the motion estimation algorithm which consumes most computational power, in terms of reducing both the frequency of block-matching and matching complexity of block-matching motion estimation algorithms and targeting high computational complexity of multi-reference frame motion estimation do research on the method of accelerating motion estimation. In view of the high computational complexity of a variety of prediction modes, this paper also explored methods of accelerating modes decision. Meanwhile, to meet the ever-changing characteristics of computing power of wireless devices, this paper also studied mechanism to make motion prediction have complexity scalability. The main results are as follows:1. A fast motion estimation algorithm based on the directional parallelogram search is proposed through in-depth studies strategies of reducing the number of block-matching to accelerate motion estimation process. First, a directional parallelogram search mode is designed by which the movement trend can be judged according to the motion vector position and directional search mode can be compatible with the trend of movement, so as to avoid blindness in the search process. Second, a search strategy with context adaptivity in terms of the search mode direction is utilized, which decides search advance direction, search mode direction, and selection of search points according to the location relationship between current optimal point and suboptimal point, such that the search path is identical to the direction of distortion decreasing, consequently obtains more motion search efficiency. In addition, a motion vector predictor called acceleration prediction motion vector is designed to enhance motion vector prediction performance. The experimental results show that the proposed fast motion estimation algorithm can get higher speedup and better rate-distortion performance than existing methods.2. A partial distortion search fast motion estimation algorithm based on initial search center predciton is proposed by study in depth methods of reducing the complexity of block-matching calculating in order to speed up the motion estimation process, which builds on the normalized partial distortion search algorithm and accelerates the convergence process of partial distortion search by means of effective initial search center prediction. An early termination detection mechanism is introduced during the search process, by which termination conditions can be judged in the phase of initial search center predciton thus subsequent unnecessary matching can be avoided. According to characteristics of the search path and distortion difference of search paths, a midway termination detection mechanism is designed, which timely judge the necessity of follow-up path search at the end of the current path search so as to further reduce search points redundancy. The experimental results show that the proposed algorithm can achieve higher speedup ratio than other similar algorithms at the same time maintaining comparable video quality. If the proposed algorithm is used combined with fast algorithm based on directional parallelogram search pattern, higher motion estimation execution speed can be obtained.3. Aiming at computational complexity linear growth brought by multi-frame motion estimation, a fast multi-frame motion estimation algorithm based on small diamond-regional selection is proposed. First, according to the characteristics of the multi-frame motion estimation, motion vector prediction mechanism in multiple reference frames is designed to improve the accuracy of the remote reference frame. Second, based on analysis of spatial and temporal distribution characteristics of the motion vector under the conditions of multi-reference-frame, a small diamond path search-based reference frame selection method is proposed, which selects the best candidate in the remaining reference frames except for the recent one. The proposed algorithm only implements complete motion estimation in two reference frame candidates and also the accuracy of frame selection can guarantee the ultimate motion prediction performance almost unaffected. The experimental results show that the proposed algorithm can significantly reduce the computational complexity of multi-frame motion estimation, while maintaining a high rate-distortion performance coherently.4. In view of the high computational complexity involved in multiple prediction modes, a hierarchical framework of prediction modes type judgment and a fast prediction mode decisioin algorithm based on this framework is proposed. According to the spatial and temporal characteristics of the macroblock represented by various prediction modes, macroblocks are hierarchically classified and therefore a suitable type of prediction mode for the current macroblock can be chosen at different classification levels based on extracted macroblock characteristics parameters. The fast mode decision algorithm is based on this framework and at different levels selects mode type according to specific extracted parameters before the complicated rate-distortion mode decision. Therefore, the choice of mode types determines prediction modes needed to be examined and that included in unchosen type can be skipped. The experimental results show that the prediction mode selection algorithm can reduce the number of prediction modes significantly meanwhile with little rate-distortion performance loss.5. In view of the ever-changing characteristics of wireless devices computing power, a motion prediction algorithm complexity scalable mechanism based on hierarchical computing power allocation is proposed According to features that implementation of motion estimation in accordance with the grid scanning, under the frame-level computing power restrictions, three level computing power allocation strategy is designed for motion prediction, including the initial, global, and local allcocation, which effectively allocates computing power to each macroblock by comprehensive utilization of the information of completed frame the intermediate results of the current frame implementation. Extensible motion estimation search patterns are used to take full advantage of the allocated computation ability if redundant so as to further improve prediction accuracy. The experimental results show that the proposed mechanism can achieve fine granularity complexity scalability for prediction algorithms and ensure the highest possible overall prediction accuracy and visual quality under computing power constraints.Integrated application of aforementioned research results, a software video encoder prototype system is designed. The prototype system is based on H.264 reference software, conducts an appropriate optimization for the code procedures and data structures, the same time integrates algorithms presented in this paper. The experimental results show that the prototype system can achieve good encoding speedup results and rate-distortion performance that further validate the the good performance of algorithms proposed by this paper.