基於空間可變之動作補償預測的動向不確定性之分析

Abstract

這份論文中提出了如何利用空間可變之動作補償濾波器來增加動作補償估測之效率。一些近年來的研究指出原本為了解決鋸齒(aliasing)現象而設計的小數點像素精度之插補濾波器，也能在某種程度上解決動向不確定性問題。實驗結果甚至指出這些濾波器某些情況在超高清晰度(UHD)的影片的編碼效率比低解析度的來得好。基於一個為了減輕動向不確定性之編碼工具，當前的作法卻有兩個主要的問題：(1)濾波器的設計仍然是為了移除鋸齒現象，(2)對於擁有不同程度動向不確定性之像素所使用之濾波器都是故定的。在這份論文中我們對於動向不確定性進行了一些實驗來提供細部的分析，同時經由實驗結果，一些重要的發現被提出，並且我們利用這些發現來提出空間可變之動作補償濾波器之設計原則。

This thesis addresses the problem of improving temporal prediction efficiency through a spatially-variant motion-compensated filtering. A few recent studies pointed out that the sub-pel interpolation filters, originally designed to address aliasing effects, can overcome the problem of motion uncertainty to some extent. Experimental results even indicated that in terms of coding efficiency, they can improve more in UHD videos than in low-resolution ones in some cases. As a coding tool for alleviating motion uncertainty, current approach however poses two major problems: (1) the filter design is still aimed at removing aliasing effects, and (2) pixels having motion uncertainty of different degrees are forced to use the same filter. In this thesis, we carried out a few experiments to provide a detailed analysis on motion uncertainty. Also, from the experimental results, some important observations were made to form guidelines for designing a spatially-variant temporal prediction filter.