根據雙圖場影像建構適應性混合式解交錯技術之研究

Abstract

在視訊解交錯的相關研究中，動作補償是許多高階的解交錯方法之 一。由於動作補償在鏡頭的移動與物體的移動時，可以補償出較好的垂直細節。而邊緣線內插法則廣泛的應用於重建影像邊緣並增加邊緣線的細膩。為保有影像移動補償以及降低動作估計所需的記憶體並減少硬體代價，本文提出以雙圖場影像的適應性混合方法的動作補償解交錯技術，本文提出的解交錯是以雙圖場的動作估計補償與改良的邊緣線內插為主體的解交錯。本文以動作估計補償為主要的架構，從前一張影像做動作估計所得的移動資訊來補償，搭配移動向量與差異比對的動作適應性的決策法則，與改良的子像素邊緣線內插方法，會使得輸出的影像具有適應性的補償。適應性的決策法則利用其移動判斷資訊做進一步的動作補償的分類，再混合適當的補償。其適應性決策法以相鄰移動向量與差異值偵測比對動作的資訊，進而改善移動向量不準確之下，所造成的視覺上的缺失。整體來說，本文提出的架構，可由適應性的決策法則再搭配動作估計補償與改良的空間性邊緣線內插得到合適的影像。其實驗結果顯示，而實際的顯示效果也比一般解交錯方法有更進一步的提升。從PSNR的比較亦可改善其輸出效能約2db，可比一般的動作補償解交錯有更好的影像顯示品質，其亦可減少多圖場動作估計的所需的硬體代價。

This thesis presents an adaptive deinterlacing algorithm that switches the deinterlacing policies between motion compensation and edge-directed interpolation. The motion- compensation-based deinterlacing is one of the most advanced approaches among deinterlacing techniques in that motion compensation can produce better quality on vertical details than the others. However, when the motion estimation is inaccurate the deinterlacing might result in blocking effects. The blocking effects can be dismissed by the edge-directed interpolation. Therefore, this thesis proposes an adaptive hybrid deinterlacing technique for two-field image reconstruction. The hybrid deinterlacing algorithm is based on two-field motion estimation and adaptive motion decision system. The adaptive motion decision system counts the variation of neighborhood motion vector and SAD to evaluate the error of motion estimation. This can help the deinterlacing to avoid from the interlace artifacts. In the edge-directed interpolation, this thesis presents the sub-pixel interpolation scheme to smooth the edge of object and reduce serration on edges. The experiment results show that the proposed algorithm can produce better video quality than conventional methods by 2 dB of PSNR while the computational complexity is low.