臺灣博碩士論文加值系統

由於訊號傳輸頻寬上的限制, 如何在整合服務數位網路及高畫質電視的影像傳輸上完成高倍率的影像壓縮, 已成為一非常重要的研究領域。在眾多的編碼技術中, 運動補償影像編碼為一高效率之編碼技術。而此編碼技術又以運動估測為其樞紐, 攸關編碼之成效。因此, 運動估測方法的發展便成為設計此編碼器的重點工作。本研究中,我們首先發展一個新的運動估測方法,並估測其在整個編碼器的地位與重要性。 基於硬體技術以及即時處理方面的考量, 區塊比對演算法是目前最具潛力的運動估測演算法。於本研究中, 我們提出一個應用降取樣法階段式區塊比對演算法及其超大型積體電路架構, 並透過巧妙的資料流將此二者整合為一有效的系統。此系統的優點可由三方面來說明 : 首先, 它的速度不亞於現有的階段式演算法。其次,它能得到良好的預測品質。再則, 本系統能克服一般階段式區塊比對法在硬體實現上所面臨的困難。我們將就目前互補式金氧半導體技術來分析系統效益。其分析結果說明, 無論是影像電話或電視傳播的畫面規格, 本系統均能於單一晶片上達成即時處理的目的。同時, 以計算機對實際影像序列作模擬的結果亦同時顯示, 所提出之演算法其估測能力接近於具最佳化的全面搜尋演算法及零等待周期演算法。

Due to the limitation of transmission bandwidth, low-bit rate motionvideo coding and transmitting for integrated service digital networks (ISDN) and high definition television (HDTV) systems has an important andgrowing research area. To realize a high-compression codec (coder and decoder), several image coding techniques have been proposed. Among these, motion compensated image coding is widely used to reduce the temporal redundancy, and the key to success of such scheme is to develop a good motion estimation technique to predict the current frame based on the previous ones. In the beginning of the thesis, a hybrid motion compensated coder is described to illustrate the importance of proposed algorithm and architecture. With regard to the computational complexity and implementation of VLSI architecture, the block matching algorithm is a good candidate for motion estimation. In this thesis, a fast hierarchical block matching algorithm and its VLSI array architectures are proposed. Based on a novel data flow, these two parts are integrated into an efficient system. The novelty of this new system is threefold. Firstly, it has high throughput.Secondly, it has good prediction quality. Thirdly, its hardware can be implemented without the limitations of the conventional hierarchical algorithms. The performance is evaluated from the viewpoint of current CMOStechnology. It shows that the real time processing of videophone and TV signals can be achieved for a single chip realization. The simulation results indicate that the prediction quality is close to the full search algorithm and zero-waiting cycle algorithm.