H.264視訊解碼器之動態補償內插架構設計與實現

Abstract

在視訊壓縮領域中，H.264/AVC是新一代的視訊壓縮標準，相較於MPEG-2，MPEG-4，和H.263，H.264/AVC提供了更高的壓縮效率以及更好的視訊品質。在視訊解碼標準中，動態補償(Motion Compensation)是其中一項提升視訊品質的技術，包括亮度部分六階有限脈衝響應濾波器和四分之一像數精確度以及彩度部分八分之一像數精確度。在整個解碼端的執行時間上動態補償佔了39% 運算時間，因為其複雜的運算，使得動態補償是影響整個系統最重要的因素。因此本篇論文的主旨在於針對降低運算複雜度設計適用於H.264/AVC視訊解碼器之動態補償內插架構。由於亮度部分佔整個動態補償運算的80%，所以針對亮度部分提出了兩種演算法來降低運算的複雜度--分別是應用六階有限脈衝響應濾波器與平均值概念之六階平均值濾波器以及應用四階有限脈衝響應濾波器與平均值概念之四階平均值濾波器。此外，針對資料供給機制我們提出了Snake Path的方法有效的降低動態補償的硬體成本，而針對所提的方法採用了平行化和管線化的架構來實現。透過軟體的模擬，我們所提的六階平均值濾波器相較於傳統六階濾波器在重建視訊品質上呈現了近似影像品質的表現。本設計架構利用TSMC 0.13 μm的製程技術合成，結果顯示我們所提出的方法比起傳統六階濾波器分別能節省48% 和68% 的硬體面積。

H.264/AVC is the newest video coding standard. Compared with MPEG-2, MPEG-4, and H.263, H.264/AVC provides highest coding efficiency and better video quality. Motion compensation is one of the techniques for enhancement video quality in video decoder standard that includes 6-tap FIR filter and 1/4-pel precision for luminance and 1/8-pel precision for chrominance. In H.264/AVC decoder, the motion compensation consumed 39% decoding time, which is the most important influence of decoding system because its complex computation. For this reason, in this thesis, we proposed efficient motion compensation interpolation architecture for lower complexity computation in H.264/AVC decoder. Since the luminance computation occupies 80% of whole motion compensation interpolation computation, consequently, we propose two algorithms to reduce the computational complexity for luminance component--the Six-Tap Mean Filter, which is apply Six-Tap FIR filter method and Mean concept, and the Four-Tap Mean Filter, which is apply Four-Tap FIR filter method and Mean concept, separately. Furthermore, we propose a novel data supply mechanism: Snake Path in order to save area. Presently, we adopt parallel and pipelined architecture for our proposed method in hardware implementation. Through the software simulation, the reconstructed video quality shows approximate image quality compare with traditional 6-tap FIR filter adopting our proposed Six-Tap Mean Filter algorithm. Our designs are synthesized with TSMC 0.13 μm technology. The synthesized results shows our propose algorithms with Snake Path control compare with traditional 6-tap FIR filter can save the hardware cost 48% and 68%, separately.