Font Size: a A A

Research On Fast Algorithm Techniques For Multiview Video Coding

Posted on:2011-04-21Degree:DoctorType:Dissertation
Country:ChinaCandidate:W ZhuFull Text:PDF
GTID:1118330332984610Subject:Electronic information technology and instrumentation
Abstract/Summary:PDF Full Text Request
Multiview video is captured by a set of cameras from different angles on the same scene. Due to the content information from different angles, it comprises rich 3D information of a scene, and it can expand the user experience beyond what is offered by the traditional media. However, with the increasing number of views, it consumes an ultra large amount of video data bandwidth. Multiview Video Coding (MVC) tries to compress multiview video data for efficient storage and transmission. With the emergence of 3D applications and the demand of real visual perception, MVC attracts more and more attention. However, the computational complexity of MVC is huge due to encoding multiple views, and it is further aggravated by employing complicated prediction structure for improving the encoding efficiency. Due to the heavy computational complexity, MVC has seriously slowed down its applications. The fast algorithms for traditional single view video coding have not considered new features of MVC, they can not fully fit the situation of MVC. Therefore, the fast algorithms of MVC should have to be studied to consider these new features. This research paper makes an attempt to address the fast algorithm techniques of MVC based on this background.In Chapter 1, the significance of the research work is presented together with a brief summary of the present research status. Then, the main research content and the structure of the thesis will be introduced.Chapter 2 proposes a fast mode decision algorithm based on the textural segmentation and correlations for MVC. Firstly, each frame is segmented into three textural regions based on the correlation between textural complexities and Intra mode Rate-Distortion (RD) costs, and the correlation of Intra mode RD costs between views. Then, by using textural region types and the correlation of Skip mode RD costs between views, an early decision of Skip mode is introduced. Thirdly, by utilizing textural region types and the correlation of prediction direction between Inter modes, the inter-view prediction of other Inter modes is selected. Finally, the estimation of Inter8×8 mode is optimized according to textural region types and motion activities.Chapter 3 presents two key elements in this thesis. A fast disparity estimation is proposed by using the spatio-temporal correlation and the temporal variation of the disparity field:Firstly, the estimated disparity vector fields of the previous coded frames are regulated and smoothed by using the global disparity vector and the spatial median filter to remove noisy disparity vectors. Then, a temporal prediction of the real disparity vector is calculated first by utilizing the smoothed disparity field of the previous coded frame. Thirdly, the search center is selected among the candidates obtained from spatio-temporal neighboring disparity vectors, and deemed to be a preliminary disparity vector. Finally, the search range is predicted adaptively by using the deviation between the search center and the temporal prediction of the real disparity vector, and then the search is implemented in a limited range; the distance represents the temporal variation of the disparity vector. Additionally in this section, a fast motion estimation algorithm is also proposed based on the correlation of motion vectors in the inter-view and spatial direction:Firstly, the estimated motion vector fields of neighboring views are filtered by the spatial median filter to remove noisy motion vectors. Then, a reference motion vector is selected from the filtered motion vector fields by using global disparity vectors. Thirdly, the search center is selected among the reference motion vector and spatial neighboring motion vectors. Finally, the search range is predicted adaptively based on the deviation between the search center and the reference motion vector.Chapter 4 proposes a computational complexity control algorithm for the hierarchical B pictures structure of MVC. Firstly, the computational complexity is allocated to different encoding layers, such as group of GOP, super frames, frames, and macroblocks. Especially, a computational complexity prediction and allocation algorithm based on the sum of absolute differences at zero vector and textural complexity is presented for macroblocks layer. Then, in the process of mode decision, the sequence of mode estimation is reordered based on the textural direction analysis and computational complexity. Finally, the computational complexity of the microblock is controlled accurately in the process of inter estimation. The proposed algorithm is further combined with the fast algorithms presented in Chapter 2 and Chapter 3 to control the computational complexity under the condition of fast algorithms.The final chapter concludes the new achievements of the whole research and the prospect of the future research.
Keywords/Search Tags:multiview video, multiview video coding, mode decision, textural segmentation, disparity estimation, motion estimation, search center, search range, computational complexity control
PDF Full Text Request
Related items