以特徵為基礎的視訊編碼位元配置結構

Abstract

近年來，在數位視訊通訊的服務，如視訊會議或串流等，已經廣泛地被使用。而由於在一些網路環境下，如ISDN、PSTN及網際網路等，受到傳輸頻寬的限制，因此如何在被限制的網路頻寬下提供使用者在觀看影片時能得到更好的視覺效果是一個重要的研究課題。在很多的研究中發現，人類對於畫面中移動的物體是比較敏感的，因此有不少研究便針對這些人眼比較容易注意的部份，在固定的目標傳輸位元率下降低背景的品質來增加前景移動物體的品質，使觀看者在這些吸引人注意的部份得到更好的視覺品質。 在這篇論文中，我們提出了一個以特徵為基礎的位元配置結構。這個結構利用不同區域的大小及動態資訊在固定的位元率下分配位元給這些區塊。另外我們也提出一個簡單而快速的物件切割演算法，利用動態資訊分解出在影片中人眼會被吸引的區域。我們在H.264的參考軟體JM9.5中實作這個架構並測試了一些影片。實驗結果證明這個方法可以增進影片中移動物體的視覺品質。

In recent years, applications of the digital video communications, such as video conferencing or streaming service are widely used. Since the channel rates of some networks, such as ISDN, PSTN and the Internet are limited, there is an active research topic on how to provide the user a better visual quality under a limited transmission rate.. In many studies, it has shown that human eye is more sensitive to the moving region inside a video scene. Therefore, some researches have concentrated on the human visual system and try to achieve an improved quality of the moving foreground quality as compared to the background to give the viewers a better visual quality on the these attracting regions. In this thesis, we propose a feature-based bit allocation scheme to distribute bits to different regions in a video scene in a constrained bit-rate by utilizing the feature of size ant motion of different regions. We also propose a simple and fast object segmentation method to extract the interesting regions by using motion information. We implement the scheme into the H.264 reference software JM 9.5 and test with some video sequences. Experimental results prove that we can improve the visual quality of the moving region.