Study On Motion Estimation And Rate Control In Video Compression

Motion estimation and rate control algorithms are both very important parts of video encoding systems. In order to remove interframe redundancy and thus to achieve high compression ratio, motion estimation algorithm is used to produce a prediction of a frame to be coded from previously coded reference frames. Quick search speed and high precision are essential to shorten coding time, heighten compression ratio and improve image quality. On the other hand, in order to transmit compressed video over communication channel, rate control algorithm is used to maintain output bit rate through modifying the encoding parameters. In addition, rate control algorithm plays an important role not only in stabilizing the coder, but also in improving the playback quality. Up to now, most motion estimation and rate control algorithms have disadvantages, such as high computational complexity and poor image quality. In order to resolve these shortcomings and meet the demands of different applications, this thesis is engaged in the study correlative to motion estimation and rate control algorithms.In general, the research work in this thesis could be summarized as below:Since most existing motion estimation algorithms have not found a good solution to the contradictory of computational complexity and image quality and they cannot be applied to real time applications and embedded systems, a hexagon-based block motion estimation search algorithm based on predictive motion vector (PMVHEXBS) is proposed. The proposed algorithm takes advantages of "stop when good enough" spirit and high speedup gain of hexagon-based search pattern (HEXBS) to find suited points quickly. In addition, modified partial distortion criterion (MPDC) is used to speed up further in the calculation of block distortion measure. Experiment results demonstrate that the proposed algorithm provides a significant reduction in terms of both computational cost and search points compared to the diamond search algorithm (DS) and predictive motion vector field adaptive search technique (PMVFAST). Moreover, the proposed algorithm achieves better quality performance than DS algorithm and only a little degradation as compared to PMVFAST.Due to the obvious image degradation of most fast search algorithms and the quite high computational complexity of existing partial distortion search algorithms, a fast efficient search algorithm-partial distortion search algorithm for block motion estimation based on predictive motion vector field (PMVPDS) is proposed to satisfy high image quality applications. PMVPDS takes advantage of the adjustable partial distortion criterion (APDC)proposed in this thesis, predictive motion vector field technique and half-stop technique to find suited points quickly. Experiment results demonstrate that PMVPDS provides about 2-7 and 0.6-3 times speedup respectively, and achieves better image quality compared to the normalized partial distortion search algorithm (NPDS) and the progressive partial distortion search algorithm (PPDS). Moreover, the image quality and search speed of proposed PMVPDS algorithm are adjustable.In order to reduce the computional complexity of region of interest (ROI) extraction technique and overcome the unscientific bit distribution of existing ROI-codec-supported rate control algorithms, a ROI-codec-supported adaptive rate control algorithm is proposed to satisfy low bit rate video systems. In this algorithm, a simple and efficient approach of extracting ROI is proposed. According to the image complexity and motion information, it distributes bits to ROI and non-ROI (NROI) respectively. As to ROI codec category, the judgment criterion of distinguishing between low and high bit rate coding category is derived, which makes the algorithm to select the rate-quantization (R-Q) model adaptively and decrease the rate control error. Experiment results demonstrate that the algorithm achieves a bit rate closer to the target, provides fewer skipped frames, and gets better objective and subjective image quality significantly compared with TMN7 and TMN8 algorithm.Since the TMN8 rate control algorithm must compute the variance of each macroblock and has high computational complexity, a low computational complexity rate control algorithm is proposed. Based on the rate distortion theory, an R-Q model based on mean absolute difference (MAD) is derived. During the encoding procedure, target bits are allocated for each frame according to the image complexity firstly. Then, the complexity and motion information of each macroblock are used to calculate its distortion weight. Lagrange theory is used to minimize the distortion and to find the optimized quantizer. Experiment results demonstrate that the proposed algorithm does not need to compute the variance and decreases the computational cost obviously compared with TMN8 algorithm. Compared with TMN7 algorithm, it achieves a bit rate closer to the target, provides fewer skipped frames, and gets the same control effect as TMN8 algorithm does. In addition, the proposed algorithm gets more stable and better image quality compared with TMN8 algorithm.