以特徵點圖形為基礎的影像切割及視差值傳遞之多視角對應方法

Feature-graph Based Image Segmentation and Disparity Propagation in Stereo Matching

林宏祥(Hong-Shang Lin)

指導教授 ： 歐陽明

共同指導教授 ： 鄭士康(Shyh-Kang Jeng)

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2011年)

https://doi.org/10.6342/NTU.2011.01406

摘要

在本篇論文中，我們提出了一個創新的多視角圖片對應技術。
首 先 ， 我 們 使 用 尺 度 不 變 特 徵 轉 換(Scale-invariant feature transform 或 SIFT)在圖片之間取得初始的稀疏對應點。然後利用我們所提出的以特徵點為基礎之圖片區塊分割方法(feature graph-based image segmentation)，找到更多的特徵點，並且改善特徵點之間的對應結果。
為了去除位於物體邊緣附近錯誤對應的特徵點，我們將特徵點作分群，並且在最小生成樹(minimum spanning tree 或 MST)建立過程中，對每個特徵點群集做切割，藉以找到錯誤對應的特徵點。最後，我們傳遞那些稀疏特徵點對應而得的視差值(disparity)給其他未找到適當對應的像素，產生視差值圖(disparity maps)。在視差值傳遞的過程中，我們會藉由以特徵點為基礎，在最小生成樹上面的區塊擴張(feature-based region growing on MST)，決定每個像素的鄰近特徵點。我們提出的對應技術可以廣泛用於不同種類的場景，而且幾乎沒有需要使用者調整或設定的參數。我們拿 Middlebury 網站上的測試資料去評估演算法的準確率， 測試結果顯示平均的錯誤像素比例大約在百分之十。

關鍵字

特徵點圖形 ； 特徵點群集 ； 視差值傳遞

並列摘要

In this paper, we propose a robust method for stereo matching. First,we use SIFT matching to generate initial sparse correspondence between images, and then use our feature graph-based image segmentation to discover more features and refine feature matching iteratively. To find outliers near objects’ boundary, we cluster the features into groups, and divide each feature group with minimum spanning tree construction. Finally, we generate disparity maps by propagating the disparity values of those sparse features to other near-by unmatched pixels. In the propagation process, we will determine each pixel’s neighbouring features via the feature-based region growing
on the minimum spanning tree. It contains two stages: (1) local feature disparity selection and (2) global propagation using energy minimization. Our method can be applied to a wide variety of cases, only few parameters need to be adjusted or specified. We evaluate our algorithm using the cases in the website Middlebury[SSZ01], and the results show that the average of bad pixels is about 10%.

並列關鍵字

feature graph ； feature group ； disparity propagation

參考文獻

[BRK10] Michael Bleyer, Carsten Rother, and Pushmeet Kohli. Surface stereo with soft segmentation. In CVPR’10, pages 1570–1577, 2010.

[BRK+11] Michael Bleyer, C Rother, P Kohli, D Scharstein, and S Sinha. Object Stereo — Joint Stereo Matching and Object Segmentation, pages 3081–3088. Number 1. 2011.

[BT98] Stan Birchfield and Carlo Tomasi. A pixel dissimilarity measure that is insensitive to image sampling. IEEE Transactions on Pattern Analysis and Machine Intelligence, 20:401–406, 1998.

[BVZ01] Yuri Boykov, Olga Veksler, and Ramin Zabih. Fast approximate energy minimization via graph cuts. IEEE Transactions on Pattern Analysis and Machine Intelligence, 23:2001, 2001.

[CM02] Dorin Comaniciu and Peter Meer. Mean shift: A robust approach toward feature space analysis. IEEE Trans. Pattern Anal. Mach. Intell., 24(5):603–619, 2002.