半自動化單視角影像2D轉3D技術

Semi-automatic 2D-TO-3D Conversion for a Single-View Image

藍美欣(Mei-Hsin Lan)

指導教授 ： 郭天穎

國立臺北科技大學/電資學院/電機工程系研究所/碩士(2013年)

摘要

隨著3D科技技術的發展，3D立體內容提供人們更真實的視覺饗宴，但現今的影像與視訊大多以單視角2D方式儲存，因此需要開發一套2D轉3D技術，其中深度估測為重建3D內容的首要工作。
本論文提出以近焦影像做3D的重建加上先前工作的戶外影像深度估測做結合，做出一個半自動單視角2D轉3D技術，讓單張影像2D轉3D系統更加的完善。
本論文藉由投影法先將影像分類為近焦影像以及戶外影像，並將戶外影像利用先前工作配置深度，我們將近焦影像透過影像分割並將相機參數以及相機模型結合算出合理數值，再透過K-mean分群的群心配置合理的深度，最後可利用介面，可讓使用者藉由分割資訊之協助，手動修正錯誤的區域深度，以得到逼近真實的深度資訊。

關鍵字

2D轉3D ； 深度估測 ； 場景分類 ； 影像分割 ； 物件分類 ； 灰暗通道 ； DIBR

並列摘要

Thanks to rapid development of 3D technology, stereoscopic 3D contents offer people a more realistic visual feast. However,since nowadays most of the existing images and videos are still stored in 2D format, it is necessary to develop a 2D to 3D technique. Depth estimation is the essential step for the reconstruction of stereoscopic 3D contents.
This thesis proposes a new method of 3D reconstruction on close-up images, combined with our previous work of depth estimation on outdoors images. Then a semi-automatic single-view 2D to 3D system is built up to make the system more complete.
This work first classifies the close-up images and outdoor images by image projection. The outdoor images are processed by our previous proposed method, whereas the close-up image is analyzed by segmentation, camera parameters and models, followed by K-mean clustering to a reasonable depth value.Finally our proposed system provides users with a mechanism to manually fix the incorrectly estimated depth based on segmentation cues to approach the real depth.

並列關鍵字

2D-to-3D conversion ； depth estimation ； scene classification ； object classification ； image segment ； dark channel ； DIBR

參考文獻

[25] L. C. Chiu and C. S. Fuh, “An efficient auto focus method for digital still camera based on focus value curve prediction model,” J. Inf. Sci. Eng. 26, 1261–1272 (2010).

[1] C. Fehn, “Depth-image-based Rendering (DIBR), Compression and Transmission for A New Approach on 3D-TV,” Proc. of the SPIE, vol. 5291, pp. 93-104, 2004.

[2] Yea-Shuan Huang, Fang-Hsuan Cheng and Yun-Hui Liang, “Creating Depth Map from 2D Scene Classification,” 3rd Int’l Conf. Innovative Computing Information and Control, pp.69, 2008.

[5] Jaeseung Ko, Manbae Kim, and Changick Kim, “2D to 3D stereoscopic conversion: depth-map estimation in a 2D single-view image,” Proc. SPIE, the International Society for Optical Engineering, vol. 6696, 2007.

[7] J. Lee, S. Yoo, C. Kim, and B. Vasudev, “Estimating Scene-Oriented Pseudo Depth With Pictorial Depth Cues,” IEEE Transactions on Broadcasting, Vol. 59, No. 2, pp. 238-250, June 2013.