| Summary: | 博士 === 國立交通大學 === 電子研究所 === 99 === Pattern-based block motion estimation (PBME) is one of the most widely adopted compression tools in the contemporary video coding systems. However, despite that many researchers have studied PBME, few have attempted to construct an analytical model that can explain the underneath principle and mechanism of various PBME algorithms.
In this dissertation, we propose a statistical PBME model that consists of two components: 1) the statistical probability distribution of the motion vectors, and 2) the minimal number of search points (called weighting function, WF) achieved by a search algorithm. We verify the accuracy of the proposed model by checking the experimental data. Then, two application examples using this model are proposed. Starting from an ideal weighting function, we devise a novel genetic rhombus pattern search (GRPS) to match the design target. Simulations show that comparing to the other popular search algorithms, GRPS reduces the average search points for more than 20% and, in the meanwhile, it maintains a similar level of coded image peak signal-to-noise ratio (PSNR) quality. Furthermore, the proposed model can reliably predict the performance of a PBME algorithm applied to a new video sequence.
With the aid of the proposed model, we design new PBMEs by looking into every component of a typical PBME algorithm and fine-tuning the major components systematically to achieve the optimal or nearly optimal results. First, we use the aforementioned analytic model in analyzing and designing effective genetic-algorithm-based pattern searches. Then, we propose an adaptive switching strategy that dynamically switches between two pattern searches. Third, we extend our PBME model to evaluate the efficiency of starting (initial search) points. A near optimal set of starting points is identified through iterative steps. Fourth, we study the early termination threshold technique and suggest a metric in selecting an effective threshold. An early termination mechanism with accurate threshold is thus constructed. Combining all these techniques, we develop a PBME algorithm that outperforms many existing algorithms.
Although the WF matches the deterministic search schemes quite well, however, the WF fails to give a precise search point prediction when a probabilistic search method such as a genetic pattern search is involved. Therefore, we propose a refined weighting function (RWF) that describes both genetic and non-genetic pattern searches more accurately under the assumption that the matching error surface is a quadrant monotonic function with smooth quadrant border (QMSB). In the process of constructing RWF, we further accelerate the pattern searches and two momentum-directed genetic pattern search algorithms are devised. These new algorithms assign priorities to the candidate mutations based on the information provided by the preceding successful searches and this can further reduce the computational complexity of the previously proposed genetic pattern searches by 5% to 7% in average.
With refined RWF, the prediction accuracy of the refined model is significantly improved. Consequently, we re-examine the coding tools in the adaptive pattern search scheme. We focus on two components, the pattern switching strategy and the starting point selection. We investigate the optimal parameter selection issue in these tools and their impacts on the overall coding performance. Experimental results show that our refined pattern switching schemes can further accelerate the search process and in the meanwhile keep the visual quality comparable to the best of their constituent pattern searches.
In summary, we propose an analytical model for PBME and demonstrate a methodology for developing new pattern-based search algorithms and the adaptive pattern search schemes by using our proposed model. One step further, we refine the original model, improve its accuracy and then design better fast search algorithms accordingly.
|
|---|
