三維立體多視角影像之空間域壓縮技術

Spatial Domain Compression Technology of 3D Stereoscopic Multiview Video

游濬箖(Jun-Lin You)

指導教授 ： 范育成

國立臺北科技大學/電資學院/電腦與通訊研究所/碩士(2010年)

摘要

由多台攝影機拍攝而得到的多視角影像(Multiview Video)可用於三維立體多視角影像的產生，多視角3D顯示器可提供多個角度讓觀看者自由地選擇，而且不須要戴上特殊的立體眼鏡，使觀看者有更自然且無限制的視覺體驗。因為多視角影像的資料量是傳統單一視角影像的二倍到數倍，用於影像儲存或傳輸時會有資料量龐大的問題。
本論文所探討的是三維立體多視角影像之空間域壓縮技術，此壓縮技術針對多視角影像資料量龐大的問題做處理，以九個視角(9-View)的影像作為研究的基礎，對多視角影像進行資料量的壓縮。在三維立體多視角影像之空間域壓縮技術中，使用視差估計(Disparity Estimation)找出相互視角(Inter-View)之間的視差向量(Disparity Vector)，再利用視差向量進行視差補償(Disparity Compensation)，補償後的影像與原參考影像相減後會獲得差異部分(Difference)，接下來對影像部分進行三維離散餘弦轉換(Three Dimensional Discrete Cosine Transform, 3D-DCT)，並且對轉換後的係數做量化，最後獲得視差向量與量化後的係數。
論文中針對3D-DCT做硬體架構的實現，可用在3D立體多視角影像壓縮方面，3D-DCT電路主要是由1D-DCT電路、SRAM、多工器和控制器所組成，其中的1D-DCT電路以移位器取代掉乘法器，可降低電路的複雜度。
本篇論文所提出的三維立體多視角影像之空間域壓縮技術具有大量縮減多視角影像資料量的優點，對未來三維立體多視角影像壓縮的發展會很有幫助。

關鍵字

多視角影像 ； 視差估計 ； 三維離散餘弦轉換

並列摘要

Multiview Video, captured by multiple cameras, can be used for the generation of 3D stereoscopic multiview video. Generated stereoscopic video, for multiview 3D displays, can provide multiple angles to choose freely for viewers and let viewers have more natural and limitless visual experience without special stereoscopic glasses. Because the amount of multiview video data is two or more times than traditional mono-view video, there will be a problem of huge amount of data for video saving or transmission.
The thesis focuses research direction of spatial domain compression technology of 3D stereoscopic multiview video. The compression technology deals with the problem of huge amount of multiview video data. The research is based on nine-view video to compress the amount of multiview video data. In the spatial domain compression technology of 3D stereoscopic multiview video, disparity estimation is used to find inter-view disparity vectors. Disparity vectors are used to perform disparity compensation. After compensated video is subtracted from original reference video, the difference can be gotten. Then, the 3D video is transfer using 3D-DCT technique. Afterward, we quantize the coefficients transformed by 3D-DCT. Finally, disparity vectors and quantized coefficients will be gotten.
3D-DCT is implemented by hardware architecture in the thesis and can be used in the compression of 3D stereoscopic multiview video. 3D-DCT circuit mainly consists of 1D-DCT circuits, SRAMs, multiplexers and a controller. It can reduce the complexity of the 3D-DCT circuit that 1D-DCT circuits use shifters to replace multipliers.
The spatial domain compression technology of 3D stereoscopic multiview video, proposed by the thesis, can reduce the amount of multiview video data greatly, and there will be a great help for the compression of 3D stereoscopic multiview video in the future.

並列關鍵字

Multiview Video ； Disparity Estimation ； 3D-DCT

參考文獻

[2] M. Zwicker, S. Yea, A. Vetro, C. Forlines, W. Matusik, and H. Pfister, “Multi-view Video Compression for 3D Displays,” Conference Record of the Forty-First Asilomar Conference on Signals, System and Computers, Nov. 2007, pp. 1506-1510.

[4] T. Fryza, “Improving Quality of Video Signals Encoded by 3D DCT Transform,” 48th International Symposium ELMAR-2006 focused on Multimedia Signal Processing and Communications, Jun. 2006, pp. 89-93.

[7] J.-W. Kang, S.-H. Cho, N.-H. Hur, C.-S. Kim, and S.-U. Lee, “Graph Theoretical Optimization of Prediction Structure in Multiview Video Coding,” IEEE International Conference on Image Processing, vol. 6, Sep.-Oct. 2007, pp. 429-432.

[8] L. Ma, and F. Pan, “Efficient Compression of Multi-View Video Using Hierarchical B Pictures,” International Conference on Multimedia and Ubiquitous Engineering, Apr. 2008, pp. 118-121.

[15] H. Pan, and F. Pan, “Development of Multi-View Video Coding Using Hierarchical B Pictures,” Congress on Image and Signal Processing, vol. 1, May 2008, pp. 497-503.