基於相機投影模型與灰暗通道模型之單視角影像2D轉3D技術

2D-to-3D Conversion for Single-View Image Based on Camera Projection Model and Dark Channel Model

林佳慶(Chia-Chin Lin)

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(2011年)

摘要

3D立體顯示技術逐漸成為未來的主流，而目前多數現存的照片皆為單視角2D格式，因此為了將其轉換為立體影像格式，本論文將針對單視角影像估測深度資訊。目前2D轉3D的文獻雖然能估測不同物件彼此的前後關係，但卻無法提供合理之相對深度；而透過統計訓練之演算法，雖然能在某些場景下逼近於真實深度資訊，但難以找到泛用性高的訓練特徵。
本論文主要針對單視角之戶外影像估測深度資訊，我們將透視投影與針孔相機模型結合，用於估測合理之相對深度，再將場景中之物件分類為四種類型，分別為天空、地面、人工建物與自然景物，並且根據物件類型採用不同的深度配置方式，最後自然景物之深度，將更進一步透過灰暗通道模型修正。實驗結果證明我們估測之深度資訊，可以在不需訓練的情況下，逼近於真實深度資訊，並且轉換之3D影像能提供令人滿意的立體感。

關鍵字

2D轉3D ； 深度估測 ； 透視投影 ； 物件分類 ； 地平線偵測 ； 灰暗通道 ； 消失點

並列摘要

Stereoscopic 3D display becomes accessible to all people. Most existing digital photos are captured in monocular 2D format, and it is desire to estimate their depth for the purpose of stereoscopic format conversion. Existing 2D-to-3D works may be able to estimate depth but they can’t provide reasonable relative depth between different objects. Approaches based on statistical training with the ground truth database may approximate some scenarios of images, but it is easy to fail with caveat that it’s hard to adopt representative training features without loss of generality.
This paper focuses on depth estimation for the monocular images taken outdoors. We combine the perspective projection with pinhole camera model with the intention of estimating reasonable relative depth between different objects. We also classify objects in scene into four types - sky, ground, man-made and natural objects; and assign depth to each type with different rules. The depth of natural objects is further corrected by dark channel model. The experiments show our estimated depth map can approximate the ground truth without training, as well as providing the satisfyingly visual results.

並列關鍵字

2D-to-3D conversion ； depth estimation ； perspective projection ； object classification ； horizon detection ； dark channel ； vanishing point

參考文獻

[2] A. Saxena, S. H. Chung, and A. Y. Ng, “3-D Depth Reconstruction from a Single Still Image,” Int. J. of Computer Vision, 2007.

[3] 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.

[4] Tao Zhang, “3D Image format identification by image difference,” IEEE Int. Conf. on Multimedia and Expo., pp. 1415-1420, 2010.

[5] 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.

[7] C. A. Chien, C. Y. Chang, J. S. Lee, J. H. Chang, and J. I. Guo, “Low Complexity 3D Depth Map Generation For Stereo Applications,” ICCE, 2011.