基於圖片切割色彩分群之立體視差圖估算

本論文根據人類心理學提出了一個基於圖片切割(segmentation-based)的創新技術。利用將各個切割區域分群的方式來幫助被因為被遮擋而分成好幾個不同區域的部分取得三維上屬於同一平面的深度估計。這樣一來就算在物體遮擋複雜的情況，我們也能得到精確的深度估算。立體視差估算(Stereo Matching)是透過模擬人類雙眼視覺來計算出圖片上物體深淺的方法。但由於一般做法上的限制使得實際上屬於同一物體或背景而在畫面上因為前景物體遮擋而被隔開的區域不容易在深度上是連續的。由於同一物體內的不同區塊通常有著類似的色彩分布，所以我們使用顏色做為各個區塊分群的基礎，使顏色相近的區域更容易被歸在同一個物體內而取得三維上的相連，透過格式塔(Gestalt)學派中的原則─接近 (proximity)的使用，和QPBO以及fusion-move等圖分割(graph-cut)技術來套用在能量最小化(energy minimization)上，以取得更好的深度估算。我們使用米德爾伯里線上立體評估(Middlebury stereo evaluation) 和它的測試資料來評估我們演算法的準確率。

關鍵字

立體視差估算；接近；圖分割

並列摘要

This paper presents a novel method according to the human perception theory. We propose a segmentation-based stereo matching method to help the discontinuous segments that belongs to the same object to have 3D connectivity. Therefore, our method can get more discriminative disparity estimation for complex occlusion. Stereo matching is a correspondence method which uses two images for simulating human eyes to do depth reconstruction. But general stereo matching methods have limited capability to retrieve the geometric surface of the object or background that is occluded and divided into many regions by other objects in front of it. In our measurement, all segments are clustered into several groups based on color cue because the color segments in the same object usually have homogeneous color. Then those color segment regions within a same color group tend to be assigned to a same object. We use the proximity principle from Gestalt psychology to properly deal with the plane labeling in each segment group. The 3D connectivity term is encoded on the proximity formulation in our proposed energy function. We also use graph-cut technique, fusion move and Quadratic pseudo-boolean optimization (QPBO), to find approximate global solution for segment plane assignment. We use CVPR 2002 dataset as source images and evaluate our result by Middlebury.

並列關鍵字

Stereo Matching ； proximity ； graph-cut

參考文獻

[1] D. Scharstein, R. Szeliski, and R. Zabih. “A taxonomy and evaluation of dense two-frame stereo correspondence algorithms.” International Journal of Computer Vision 47, 2002, 7–42

[3] Oliver Woodford, Philip Torr, Ian Reid, and Andrew Fitzgibbon. “Global stereo reconstruction under second-order smoothness priors.” In Pattern Analysis and Machine Intelligence, IEEE Transactions on, 31(12):2115–2128, 2009.

[5] A. Klaus, M. Sormann, and K. Karner. “Segment-based stereo matching using belief propagation and a self-adapting dissimilarity measure.” In Proceedings of International Conference on Pattern Recognition, 2006.

[8] Victor Lempitsky, Carsten Rother, “Fusion Moves for Markov Random Field Optimization” In Pattern Analysis and Machine Intelligence, IEEE Transactions on 32.8 (2010): 1392-1405.

[9] Vladimir, Kolmogorov, and Carsten Rother,” Minimizing non-submodular functions with graph cuts – a review” In Pattern Analysis and Machine Intelligence, IEEE Transactions on 29.7 (2007): 1274-1279.

國際替代計量

基於圖片切割色彩分群之立體視差圖估算

全文下載

主題瀏覽