| With rapid development and extensive application for digital video, traditional two-dimensional images has been unable to satisfy people's demand for video quality and content diversity. As the next generation video display technique after high-definition television (HDTV), three-dimensional television (3DTV)/free viewpoint television (FTV) is considered as a promising multimedia service of next generation. However a large amount of date is extremely tough to transmit and store multi-view video data, which are the bottleneck problem in the development of multi-view video technologies. Therefore, multi-view video coding has been intensively studied in recent years.This dissertation investigates H.264/AVC based multi-view video coding algorithm and related techniques. The major contents and novelties of this thesis are summarized as follows:Joint Multi-view Video Coding (JMVC) proposed the fixed view-temporal prediction structure for the standardization multi-view video coding (MVC), which is difficult to cope with various characteristics of multi-view videos. This paper proposes a new inter-view prediction structure, which determine the location of I-view based on correlation analysis, and determine numbers of B-frames between I-(or P-) and P-frame considering coding efficiency and random access and dynamically select to the inter-view prediction structure. Experimental results show that the proposed prediction structure can improve coding efficiency.To display video on multitude of different terminals, a multi-view coding scheme with view scalability (SMVC) is proposed. Based on a correlation analysis, all views are divided into two groups:based view and enhancement views. The proposed SMVC adjusts characteristics of various multi-view videos. Experimental results show that the scheme determines its coding efficiency and view scalability. In order to satisfy transmission and application, this paper illustrates the drawback of current existed rate-distortion model and the characteristic of multi-view video coding, and then proposes frame-layer and macroblock-layer rate control algorithm for multi-view video coding based on the quadratic rate-distortion model. Firstly, frame-layer bit allocation for multi-view video coding according to the previously coded information keep high coding efficiency compared to the multi-view video coding with fixed quantization parameter, but the rate control error (RCE) is too large. Secondly, macroblock-layer rate control for MVC, which allocates the bits and controls the rate for view layer, frame layer, and basic unit layer, of which the view layer allocates bits according to the previously coded information and the frame layer allocates bits according to frame complexity and temporal activity. Simulation results show that the proposed algorithm can effectively control bit rate for multi-view video coding while keeping high coding efficiency compared to the multi-view video coding with fixed quantization parameter.Based on the research above, this paper classifies each picture into six frame types based on the relation between disparity prediction and temporal prediction estimation and also improves the quadratic rate-distortion model and propose frame-layer and macroblock-layer rate control algorithm for MVC based on correlation analysis. Firstly, frame-layer bit allocation for multi-view video coding which mainly allocate bit-rate among views based on correlation analysis. Compared to the multi-view video coding with fixed quantization parameter, the proposed scheme achieves up to 0.05 dB improvements in PSNR, but it controls the bit rate with an average rate control error of 1.7%. Secondly, this paper proposes macroblock-layer rate control for MVC. The proposed algorithm reasonably allocates bit-rate among views based on correlation analysis. Then the proposed algorithm consists of four levels for rate bits control more accurately, of which the frame layer allocates bits according to hierarchically structured B-frames and frame complexity and temporal activity and basic unit layer use different quantization parameters to encode various macroblocks according to the content complexity. Compared to the multi-view video coding with fixed quantization parameter, the proposed scheme achieves up to 0.18 dB improvements in PSNR and efficiently controls the bit rate with an average rate control error of 0.8%.This work is sponsored by key project of Natural Science Foundation of China《Multi-view video coding for free view video, basic theory research on 3D display and its key technologies》(Grant NO.60832003), and the project of Natural Science Foundation of China《Multi-view video coding and rendering virtual views at decoder》(Grant NO.60672052). This work is also supported by Innovation Foundation of Shanghai University《Rate control algorithm for multi-view video coding》(Grant NO.SHUCX091061).
|
PDF Full Text Request