視訊系統轉換演算法之研究與設計

Abstract

在本文中，我們針對影像格式轉換的三個領域提出數種低複雜度，且能提 升畫質的轉換理論。對於間次掃描與非間次掃描轉換，我們改良傳統的三 點中值濾波理論為動適應三點中值濾波理論，使其畫質顯著提高。而由於 掃描線數轉換需要大量的內插運算，我們也提出了新的快速內插理論，包 括一維與二維，線性與非線性內插理論推導及其硬體架構設計。可將運算 複雜度降低至一半左右且具有最低的捨入誤差，是目前最有效率的內插方 式。在圖場頻率轉換部分，基於現有的移動補償內插理論，我們提出移動 向量預估的觀念。對於降低移動向量偵測的運算複雜度及改善內插圖場的 畫質提供了相當大的助益。 In the thesis, we proposed several efficient algorithms which can improve image quality with low computational comp- lexity for three key issues in video format conversion. For interlaced- to-progressive conversion, a motion adaptive 3 point median filtering algorithm is proposed to enhance the image quality of traditional methods while still allowing cost efficient production. Also we proposed a novel and fast interpoltion and decimation algorithm which reduce the number of multiplications by nearly 50% of polyphase interpolator for scanning line conversion. It is well extended from 1-D to 2-D interpolation algorithm and hardware design which keep the same round-off noise as that of polyphase interpolator. It is most efficient algorithm we can get so far. Finally, in the field rate conversion, based on the motion-compensated interp- olation algorithm, the concept of predicted motion vector is proposed. It great reduces the computational complexity of motion estimation and achieves high quality pictures ofint- erpolated images as well.