應用於物件切割之多視角影像深度產生

Depth Map Generation Based on Video Segmentation of Multi-view Video System

簡瑋倫(Wei-Lun Chien)

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(2009年)

摘要

以目前3D領域中，多數研究領域是以利用2D資訊來產生3D立體效果為研究方向，其中一個研究領域為利用multiview video system來產生深度影像，使著可以透過深度影像來產生多視角的立體效果。以2D視訊標準來說，MPEG-4視訊標準有應用到物件切割演算法，但於3D領域中，物件切割演算法尚未廣泛的被應用，所以希望未來能夠利用物件切割搭配3D深度影像處理，以應用在不同的商業領域中。
本論文所提出的架構是利用物件切割搭配深度預測來實現物件深度效果。在物件切割方法是採用分水嶺演算法，目的是為了讓切割效果可靠度高，並且採用預測分水嶺演算法代替傳統分水嶺演算法，即可增加系統處理速度，又能保持切割的完整性與可靠度。在深度預測方面，本論文是採用三台平行式架構的相機來當作產生中間深度影像的來源，所使用的預測深度方式為，將中間影像當做current view，左右兩台影像當做reference view，利用左右兩台資訊來預測出中間影像的深度值，cost function為SAD(sum of absolute difference)；同時本系統也考慮了三種誤差因素：遮蔽效應(occlusion)、亮度差異以及影像一致性(homogeneous)，進而提升深度精確度。
本論文最重要的目的為：第一是希望能用16×16區塊大小來將預測的可靠度提高，以產生精準的深度效果；第二是將多視角可能發生的效應誤差降低，以產生精準度高的深度效果；第三是將產生的深度圖取代掉左右兩台影像資訊，進而達到壓縮目的；第四則是搭配物件切割演算法，以降低背景所產生的預測誤差，使得物件邊緣及物件本身的深度精確性提高，同時大量降低了在預測時所需花的運算量，也使得立體效果更為顯著。

關鍵字

物件切割 ； Depth Map Generation

並列摘要

In the 3D video domain, most researches are applied 2D image information to generate stereoscopic. One of researches is generating depth map of multiview video system. To apply depth map can generate multiview stereoscopic. For 2D video standard, the MPEG4 video standard includes video segmentation algorithm, but not popular in the 3D domain. In the future, the video segmentation and depth map generation can apply to different business domain.
In this thesis, the architecture is using video segmentation and disparity estimation to generate depth map. The watershed algorithm can produce high reliability in video segmentation, but the predictive watershed algorithm can promote processing speed. So the predict watershed algorithm is replaced to watershed algorithm. In the disparity estimation, the three cameras are used to produce depth map and using SAD to estimate the cost function. The system also considers three effects in order to raise accuracy：occlusion、luminance difference and homogeneous. The thesis proposed a depth map generation scheme to produce high reliability with 16×16 block size and to produce high accuracy image by reducing the effects. At the same time, the proposed method compress data by depth map and produce clear object boundary and object accuracy by video segmentation.

並列關鍵字

Video Segmentation ； Depth Map Generation

參考文獻

[1] T. Sikora, “The MPEG-4 Video Standard Verification Model,” IEEE Trans. Circuits Syst. Video Technol., vol. 7, no. 2, pp. 19–31, Feb. 1997.

[2] L. Vincent and P. Soille, “Watersheds In Digital Spaces: An Efficient Algorithm Based on Immersion Simulations,” IEEE Trans. Pattern Anal. Machine Intell., vol. 13, no. 6, pp. 583–598, Jun. 1991.

[4] S.-Y. Chien, Y.-W. Huang and L.-G. Chen, “Predictive Watershed: A Fast Watershed Algorithm for Video Segmentation,” IEEE Tran. Circuits and Systems for Video Technology, vol. 13, no. 5, pp. 453-461, May. 2003.

[5] S.-Y. Chien, S.-Y. Ma and L.-G. Chen, “Efficient Moving Object Segmentation Algorithm Using Background Registration Technique,” IEEE Trans. Circuits Syst. Video Technol., vol. 12, no. 7, pp. 577–586, Jul. 2002.

[6] D. Wang, “Unsupervised Video Segmentation Based on Watersheds and Temporal Tracking,” IEEE Trans. Circuits Syst. Video Technol., vol. 8, no. 9, pp. 539–546, Sep. 1998.