提升幀播放率- 使用基於邊緣檢測的移動補償法

Abstract

在這篇論文裡的主要目標是增加幀播放率，例如從每秒30個畫面提升成每秒60個畫面， 我們提出使用邊緣檢測演算法與移動向量的中值篩檢法, 用以強化移動向量的精確性。 利用更精確的移動向量，在相鄰兩張Frame中找出更精確的物件移動軌跡，再藉由此物件的移動軌跡， 配合所提出的內插法， 內插出和原始 Frame 相似度很高的 Frame。 我們所提的方法稱為Edge-based Motion-Compensated Frame Rate Up-Conversion，其中包含了兩個重要的部分 : 「移動向量預測」 與 「移動補償」。在移動向量預測中，我們使用邊緣偵測(edge detection) 作法中廣為人知的 SOBEL ALGORITHM 來偵測Frame中Object的邊緣，使用此邊緣得以強化移動向量的正確性。 再將所得的移動向量，經由中值篩檢法，得到更精確的修正。 實驗結果顯示我們的方法能有效的內插出 Frame，並得到不錯的 PSNR 值 。

In this thesis, a method of frame-rate up-conversion was proposed. We use the Edge Detection Algorithm and Median Vector Filter to get improved motion vector. With improved motion vectors , more accurate objects motion trajectories between two adjacent frames could be obtained. By using the proposed interpolation algorithm , the frame between its two adjacent frames could be well interpolated along the motion trajectories. The method we proposed is called Edge-based Motion-Compensated Frame Rate Up-Conversion (EBMC-FRUC) , which includes two key parts : motion estimation(MV) and motion-compensated frame interpolation(MCFI) . In motion estimation , we proposed to use Sobel algorithm which is a well known edge detection algorithm to extract the object’s edge in frame , and then use the extracted edges and block matching algorithm to obtain improved the motion vectors . The obtained motion vectors are further refined by a median filter before they are adopted in the frame interpolation. Experimental results show that the proposed algorithm effectively enhanced the correctness of motion vectors and interpolated the frames with good quality .