Fast Mode Prediction and Inter Prediction for H.264/AVC Multi-view Extension

碩士 === 國立雲林科技大學 === 電子與光電工程研究所碩士班 === 100 === Multiview video coding (MVC) has decisive effect on three-dimensional (3-D) video technology. In order to reduce its extremely high data bandwidth, MVC employs full mode decision algorithm. And not only the motion estimation (ME) based on H.264/AVC, it a...

Full description

Bibliographic Details
Main Authors: Huan-wei Fang, 房煥偉
Other Authors: none
Format: Others
Language:zh-TW
Published: 2012
Online Access:http://ndltd.ncl.edu.tw/handle/76694718837995865032
Description
Summary:碩士 === 國立雲林科技大學 === 電子與光電工程研究所碩士班 === 100 === Multiview video coding (MVC) has decisive effect on three-dimensional (3-D) video technology. In order to reduce its extremely high data bandwidth, MVC employs full mode decision algorithm. And not only the motion estimation (ME) based on H.264/AVC, it also uses high correlation of the adjacent view to add disparity estimation (DE) in each macroblock which effectively boosts efficiency of coding. On the other hand, since MVC processes additional calculation, it requires even higher calculation efficiency and memory bandwidth than H.264/AVC does, which makes it more difficult to apply it to real-time system. Therefore, this study proposes fast algorithm applicable to MVC in order to lessen calculation complexity effectively, including fast mode prediction and fast disparity estimation algorithm. In fast mode prediction, conditions are determined by coordinating the correlation of the rate-distortion cost (RD cost) between inter-views with the RD cost of the coded mode so as to make early termination for mode decision, in order to save computational complexity of mode decision. Besides, in fast disparity estimation, direction and size of disparity vector is predicted based on the disparity vector’s temporal and spatial correlation, in order to effectively reduce search range of disparity estimation and so avoid redundant calculation. Analysis of the experiment supports the fact that the above two fast algorithm integrates the inter-view and the temporal and spatial correlation, and they both created highly accurate predicted results respectively.The experimental result from our study shows that, when there is almost no peak signal-to-noise ratio (PSNR) decrease (0.009 dB in average) and only negligible increase in average bit-rate, computational complexity of the proposed algorithm is 83.08%~90.37% less than that of the fast search algorithm of MVC.