應用於視頻上的紋理預測之演算法及硬體架構實作

Algorithm and Architecture of Texture Prediction in Video Application

陳冠宇(Kuan-Yu Chen)

指導教授 ： 陳良基

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2011年)

https://doi.org/10.6342/NTU.2011.10683

摘要

無資料

關鍵字

紋理預測

並列摘要

Advanced video applications are the epochal impacts to the history of human visual perception system. The evolution of television technology grows toward more colorful and higher resolution with these applications. To pursue higher visual quality and more realistic visual perception, more and more advance video applications are developed such as high definition TV (HDTV), 3D device, Internet video streaming and free-viewpoint 3DTV/Virtual reality. In these advance video application, the main challenge is the massive data capacity and data loss. Therefore, the role of texture prediction becomes more and more significant. Many texture prediction methods have been proposed and is a well-developed technique, such as intra prediction in video coding, interpolation, inpainting, etc. In this thesis, the texture prediction algorithm and its hardware architecture design are applied in many advanced video application to further ameliorate the visual quality.

First, the texture prediction is employed as the occlusion recovery method in virtual view synthesis in multi-view system. To fulfill the more realistic experience, multi-view video brings the viewers a three-dimensional and real perceptual vision by transmitting different video sequences simultaneously on the display. However, multi-view video format is not enough to support free viewpoint sequences since its samples of spatial dimension is finite. For this purpose, the virtual view synthesis algorithm is developed for rendering images seen from any virtual viewpoints by the finite source of images seen from some fixed viewpoints only. In virtual view synthesis, the occlusion region which is blocked by the objects in reference view deteriorates the visual quality. Therefore, we propose a single iterative hybrid motion and depth-oriented inpainting algorithm and its corresponding hardware architecture to retrieve the texture in occlusion region. The simulation result outperforms by both perceptual quality and the objective metric measure. Our hardware architecture reduces 93.3% of computation cycles and still maintains the quality by isophote line propagation and depth enhancement.

Video technology contributes a lot in modern society and digitization of video further simplifies the processing, transmission and storage of video content. Without unlimited storage capacity and transmission rate, video coding is necessary. In the second part of the thesis, we introduce and analyze the newest video coding standard, high efficiency video coding. HEVC targets to further reduce the bit rate by 50% compared to the H.264/AVC, current state-of-art of video coding. With the bandwidth limitation and targeting higher resolution, texture prediction is required to reaching better coding efficiency. In this thesis, a texture prediction technique, intra plus inpainting mode, is proposed to further decrease the bit rate or lower the computation complexity. Based on the proposed algorithm and architecture, a worldwide first HEVC standard of intra prediction mode with the specification of Quad-HD 4096x2160 sequence with 30 fps is revealed.

並列關鍵字

Texture Prediction

參考文獻

ISO/IEC JTC1/SC29/WG11, N5877, July 2003.

[2] T. K. Robert T. Collins, Omead Amid, “An active camera system for

“Multipoint measuring system for video and sound - 100-camera

phy. Geometry and Computing, 2010.

[6] “Introduction to multi-view video coding,” in ISO/IEC JTC1/SC