物件輪廓導向之低位元率視訊編碼研究及其應用; Object-Based Very Low Bit-Rate Video Coding and Its Application

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/8391

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/8391

Title:	物件輪廓導向之低位元率視訊編碼研究及其應用;Object-Based Very Low Bit-Rate Video Coding and Its Application
Authors:	黃志雄;Chih-Shoung Huang
Contributors:	資訊工程研究所
Keywords:	直線霍夫轉換;霍夫轉換;非旋轉位移變化的;視訊編碼;低位元率;物件輪廓;非對稱式用戶迴路;HT;STIRS;Video Coding;very low bit-rate;VLB;Object-based;SLHT;ADSL
Date:	2000-06-22
Issue Date:	2009-09-22 11:25:44 (UTC+8)
Publisher:	國立中央大學圖書館
Abstract:	二十一世紀無疑地是網際網路與無線通訊的時代，雖然在提供寬頻的網路是未來的趨勢，頻寬的有效利用總是需要的。因為在有限的頻寬內，提供更豐富服務的需求永遠存在，因此考量視訊傳送品質與頻寬的有效利用，超低位元率視訊壓縮編碼在視訊通訊中有其必要；這類的視訊通訊包括無線視訊通訊及各種網際網路多媒體服務。超低位元率視訊壓縮編碼技術已研究長達十年，相關的國際標準有H.263, MPEG-4及MPEG-7。傳統以方塊編碼為基礎的壓縮編碼技術仍然是H.263及MPEG-4的主要編碼技術，即使「視訊物件層面」(video object plane─VOP) 已經成為MPEG-4編碼之基本元件，常用的超低位元率視訊壓縮編碼技術有兩種:模型導向編碼與物件輪廓導向編碼。近年來，由於對任意場景物件模型的難以建立以及高計算複雜度，使得物件輪廓導向編碼法則漸漸成為超低位元率視訊壓縮編碼的主流，在物件輪廓導向編碼中，必須用到三組參數，那就是物件的形狀、運動參數及色彩。他們分別由物件切割、運動估測以及運動補償三個基本單元來處理，在過去，大部份的超低位元率視訊壓縮編碼研究中，物件形狀資訊是一定要傳送到解碼端且都利用一般的區域成長法則來產生物件形狀輪廓，在參數抽取部份，利用物件在三度空間模型化的運動轉化為平面空間的運動參數估測是最近常用的法則，然通常都忽略了非剛性物件引起的失真補償。本篇論文提出一種以影像柱狀圖波谷分析以及利用物件與鄰近物件間的屬性關係，精確地切割出物件輪廓，並藉由物件區域的面積重心以及修正過的直線霍夫轉換法則穩定地抽取出物件運動參數，其中物件的形狀資訊是毋需傳送到解碼端的。本篇論文利用一個簡化的直線霍夫轉換法則大幅度的改善霍夫轉換中計算複雜度問題，同時物件方位也因此而更穩健地取得，用以估測物件移動的參數之選擇與抽取對運動估測的效能表現是非常重要的，本篇論文針對幾何參數有深入的分析與評估，另外，非剛性物件的運動補償也一併納入探討，其中利用單一碼簿的向量量化法則於一般的影像序列，而結合漢米爾編碼之混合式方塊切割編碼則適用於影像序列間差異極大的場合，本篇論文包括了完整的理論分析、實驗結果與實務應用，從實驗結果顯示，一個在物件區域切割、運動估測及估測失真補償各方面都有創意性的強化與具體的改善已經達成，再者，從實際的系統試用評估也顯示，一個高效能的實用化超低位元率視訊壓縮編碼視訊傳送系統已成功地建立。 It is true that the 21-th century is an Internet and wireless world. In spite that provides the broadband network is the trend in future, the bandwidth is effectively employed is necessary because the requirement to provide more services in a finite bandwidth is existed forever. Consider the video quality and effective usage of bandwidth, very low bit-rate (VLB) video coding is essential in visual communications, such as mobile visual communication and multimedia services on the Internet. VLB coding techniques have been developed for decade year. The related standards have H.263, MPEG-4 and MPEG-7. The traditional block—based coding is still the major technique for H.263 and MPEG-4 even video object plane (VOP) is a basic element in MPEG-4.The popular algorithms for VLB coding have two kinds. They are model-based coding and region contour-based coding. Recently the region contour-based coding becomes the major algorithm for VLB coding because of hard to establish the object model and the computational complexity for arbitrary scenes. Three parameter sets have to be used in region contour-based VLB coding. They are regional shape, motion and color and are processed by region segmentation, motion estimation and compensation. The shape information has to be transmitted and be generated by the simple region-growing method in most of past algorithms in VLB coding. The parameter modeling from 3D mapping onto 2D for object moving is mostly adopted in the feature extraction and the motion compensation of non-rigid object is neglected. A new region contour-based video coding technique is proposed in this dissertation. It precisely extracts the regions using histogram concavity analysis of the image and the attributes among the regions, extracts the motion parameters by means of region area, the gravity center of region and the modified straight line Hough transform (SLHT) algorithm. No shape information needs to be transferred to the decoder here. The huge amount of the computational complexity in the Hough transform is largely reduced by the simplified SLHT. Furthermore, the object orientation can be obtained robustly using the modified SLHT algorithm. The selection and extraction of the parameters used to estimate the object moving is very important for the performance of motion estimation. In this dissertation, the analysis and evaluation in detail for the geometric features is presented. Furthermore, the motion compensation of non-rigid objects is included in this dissertation. Meanwhile vector quantization (VQ) with a universal codebook is adopted for the general video sequences. The hybrid block truncation coding (BTC) with Hamming codes is employed to compensate the distortion caused by the non-rigid objects while the differences among the video sequences is large. The experimental results even the empirical application is presented in this dissertation. The experimental results indicate that the creative enhancements and significant improvements in region segmentation, motion estimation and prediction compensation is achieved. Furthermore, a realized VLB video delivery system with high performance can be successfully built up from the evaluation of the field trial.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Size	Format

社群 sharing

Loading...