An efficient Earth Mover's Distance algorithm for robust histogram comparison

被引:286
作者
Ling, Haibin
Okada, Kazunori
机构
[1] Univ Maryland, Dept Comp Sci, College Pk, MD 20742 USA
[2] Univ Maryland, UMIACS, College Pk, MD 20742 USA
[3] San Francisco State Univ, Dept Comp Sci, San Francisco, CA 94132 USA
基金
美国国家科学基金会;
关键词
Earth Mover's Distance; transportation problem; histogram-based descriptor; SIFT; shape context; spin image; shape matching; interest point matching;
D O I
10.1109/TPAMI.2007.1058
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We propose EMD-L-1: a fast and exact algorithm for computing the Earth Mover's Distance ( EMD) between a pair of histograms. The efficiency of the new algorithm enables its application to problems that were previously prohibitive due to high time complexities. The proposed EMD-L-1 significantly simplifies the original linear programming formulation of EMD. Exploiting the L-1 metric structure, the number of unknown variables in EMD-L-1 is reduced to O(N) from O(N-2) of the original EMD for a histogram with N bins. In addition, the number of constraints is reduced by half and the objective function of the linear program is simplified. Formally, without any approximation, we prove that the EMD-L-1 formulation is equivalent to the original EMD with a L-1 ground distance. To perform the EMD-L-1 computation, we propose an efficient tree-based algorithm, Tree-EMD. Tree-EMD exploits the fact that a basic feasible solution of the simplex algorithm-based solver forms a spanning tree when we interpret EMD-L-1 as a network flow optimization problem. We empirically show that this new algorithm has an average time complexity of O(N-2), which significantly improves the best reported supercubic complexity of the original EMD. The accuracy of the proposed methods is evaluated by experiments for two computation-intensive problems: shape recognition and interest point matching using multidimensional histogram-based local features. For shape recognition, EMD-L-1 is applied to compare shape contexts on the widely tested MPEG7 shape data set, as well as an articulated shape data set. For interest point matching, SIFT, shape context and spin image are tested on both synthetic and real image pairs with large geometrical deformation, illumination change, and heavy intensity noise. The results demonstrate that our EMD-L-1-based solutions outperform previously reported state-of-the-art features and distance measures in solving the two tasks.
引用
收藏
页码:840 / 853
页数:14
相关论文
共 51 条
[1]  
Ahuja RK, 1993, NETWORK FLOWS THEORY
[2]   A novel vector-based approach to color image retrieval using a vector angular-based distance measure [J].
Androutsos, D ;
Plataniotis, KN ;
Venetsanopoulos, AN .
COMPUTER VISION AND IMAGE UNDERSTANDING, 1999, 75 (1-2) :46-58
[3]   Shape matching and object recognition using shape contexts [J].
Belongie, S ;
Malik, J ;
Puzicha, J .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 2002, 24 (04) :509-522
[4]  
Cohen S., 1999, Proceedings of the Seventh IEEE International Conference on Computer Vision, P1076, DOI 10.1109/ICCV.1999.790393
[5]  
Darabiha A, 2003, PROC CVPR IEEE, P203
[6]   Learning low-level vision [J].
Freeman, WT ;
Pasztor, EC ;
Carmichael, OT .
INTERNATIONAL JOURNAL OF COMPUTER VISION, 2000, 40 (01) :25-47
[7]  
Grauman K, 2005, IEEE I CONF COMP VIS, P1458
[8]  
Grauman K, 2004, PROC CVPR IEEE, P220
[9]   Distance sets for shape filters and shape recognition [J].
Grigorescu, C ;
Petkov, N .
IEEE TRANSACTIONS ON IMAGE PROCESSING, 2003, 12 (10) :1274-1286
[10]   EFFICIENT COLOR HISTOGRAM INDEXING FOR QUADRATIC FORM DISTANCE FUNCTIONS [J].
HAFNER, J ;
SAWHNEY, HS ;
EQUITZ, W ;
FLICKNER, M ;
NIBLACK, W .
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, 1995, 17 (07) :729-736