以幾何運算與深度資訊傳遞為基礎之液晶鏡頭焦距可調式相機2D影像轉換至3D深度圖之研究

2D to 3D Conversion Using Camera with Liquid-Crystal Lens by Geometric Approach and Depth Map Propagation

吳宜瑾(Yi-Chin Wu)

指導教授 ： 陳永昌

國立清華大學/電機資訊學院/電機工程學系/碩士(2009年)

摘要

在立體(三維)視覺技術中，如何在拍攝的場景的同時取得場景中的深度資訊一直是一個值得深入探討的問題。雖然既有的攝影技術或是感測元件已可以在拍攝場景的同時取得場景深度，但是並不是用於目前大多數的可攜式系統，諸如手機相機或是筆記型電腦等強調低耗能需求的系統中。聚晶光電所研發的液晶變焦鏡頭提供了在低耗能的變焦相機系統，藉由此系統拍攝的二維影像與相機所提供的焦距線索提供了在低耗能需求下取得深度資訊的解決方案。
藉由焦距線索來重建二維影像中的三維資訊可分成聚焦線索方法(Depth From Focusing)與失焦線索方法(Depth From Defocusing)。DFF較精確，但需長時間對靜止場景拍攝大量二維影像；而DFD利用較少的二維影像來還原場景深度資訊，但相對的需要更多的計算需求。
在本論文中分析了液晶變焦鏡頭相機系統的特性，並且在現有的諸多深度估測的演算法中選擇了DFD方法以降低二維影像的拍攝張數需求，其中又選擇幾何方法來實現原始深度資訊的計算與估測，以期降低液晶鏡頭相機系統拍攝影像中的雜訊影響。而針對沒有紋理的區域，也就是無法提供任何失焦線索的區域，本論文選擇了最佳化方法，將紋理區域上可靠的深度估測推演至沒有紋理的區域。
本論文使用基於深度圖之立體表現系統(Depth-Image-Based Rendering System)之立體顯示器來驗證實驗結果，並且得到令人滿意的立體視覺效果。

關鍵字

深度圖估算 ； 二維至三維轉換 ； 由失焦產生深度 ； 影像處理 ； 三維場景重建

並列摘要

In the development of 3D visualization techniques, how to acquire an image along with its depth information is always one of the most interesting problems worthy of research. In order to obtain a 3D image with its depth information in a scene from the 2D images captured from an ordinary camera system, it may not be feasible with the portable devices such as mobile phones and laptops which emphasize low power consumption. The Liquid-Crystal Lens camera developed by Tunable Optix Inc. may provide a new approach to capture the 2D images with low power consumption and also provide the focal cue to reconstruct the depth.
There are two main approaches for 3D reconstruction from 2D images’ focal cues. One approach is Depth from Focusing (DFF) and the other is Depth from Defocusing (DFD). DFF has more accuracy, but has to take much more images of a stationary scene. DFD needs fewer images to reconstruct the depth but is more computation expensive.
In this thesis, we analyze the characteristics of the LC lens camera system and chose the geometric approach to implement the initial depth estimation. The geometric approach is a DFF method and would be more robust against the intensive noise caused by the LC lens camera system. And then we employ a depth propagation process using optimization to refine the depth map in plain regions of a 2D image.
In the experiment, our method can generate rather good 3D visualization results, verified by a DIBR 3D display system.

並列關鍵字

Depth map estimation ； 2D to 3D conversion ； Depth from defocusing ； Image processing ； 3D reconstuction

參考文獻

[1] C. Fehn, R. Barre and R. S. Pastoor, “Interactive 3-D TV—Concepts and key technologies,” Proc. IEEE, vol. 94, pp. 524, Mar. 2006.

[2] Nayar, S.K., ”Shape from Focus System” Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Champaign, Illinois, pp.302-308, June 1992.

[5] Elder J. H. and Zucker S. W., “Local scale control for edge detection and blur estimation.”, IEEE Transactions on PAMI, vol. 20, NO. 7, , pp. 699–716, July 1998

[6] M. Subbarao and G. Surya, “Depth from Defocus: A Spatial Domain Approach,” Int'l J. Computer Vision, vol. 13, pp. 271-294, 1994.

[3] Murali Subbarao, Tae Choi and Arman Nikzad,” Focusing Techniques”, Journal of Optical Engineering, vol. 32, pp. 2824-2836, Nov.. 1992.