Alexander Berg, Tamara L. Berg and Jitendra Malik

EECS Department
University of California, Berkeley
Technical Report No. UCB/CSD-04-1366
December 2004

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2004/CSD-04-1366.pdf

We approach recognition in the framework of deformable shape matching, relying on a new algorithm for finding correspondences between feature points. This algorithm sets up correspondence as an integer quadratic programming problem, where the cost function has terms based on similarity of corresponding geometric blur point descriptors as well as the geometric distortion between pairs of corresponding feature points. The algorithm handles outliers, and thus enables matching of exemplars to query images in the presence of occlusion and clutter. Given the correspondences, we estimate an aligning transform, typically a regularized thin plate spline, resulting in a dense correspondence between the two shapes. Object recognition is then handled in a nearest neighbor framework where the distance between exemplar and query is the matching cost between corresponding points. We show results on two datasets. One is the Caltech 101 dataset (Fei-Fei, Fergus and Perona), an extremely challenging dataset with large intraclass variation. Our approach yields a 48% correct classification rate, compared to Fei-Fei et al.'s 16%. We also show results for localizing frontal and profile faces that are comparable to special purpose approaches tuned to this task.

BibTeX citation:

@techreport{Berg:CSD-04-1366,
    Author = {Berg, Alexander and Berg, Tamara L. and Malik, Jitendra},
    Title = {Shape Matching and Object Recognition using Low Distortion Correspondences},
    Institution = {EECS Department, University of California, Berkeley},
    Year = {2004},
    Month = {Dec},
    URL = {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2004/6200.html},
    Number = {UCB/CSD-04-1366},
    Abstract = {We approach recognition in the framework of deformable shape matching, relying on a new algorithm for finding correspondences between feature points. This algorithm sets up correspondence as an integer quadratic programming problem, where the cost function has terms based on similarity of corresponding geometric blur point descriptors as well as the geometric distortion between pairs of corresponding feature points. The algorithm handles outliers, and thus enables matching of exemplars to query images in the presence of occlusion and clutter. Given the correspondences, we estimate an aligning transform, typically a regularized thin plate spline, resulting in a dense correspondence between the two shapes. Object recognition is then handled in a nearest neighbor framework where the distance between exemplar and query is the matching cost between corresponding points. We show results on two datasets. One is the Caltech 101 dataset (Fei-Fei, Fergus and Perona), an extremely challenging dataset with large intraclass variation. Our approach yields a 48% correct classification rate, compared to Fei-Fei et al.'s 16%. We also show results for localizing frontal and profile faces that are comparable to special purpose approaches tuned to this task.}
}

EndNote citation:

%0 Report
%A Berg, Alexander
%A Berg, Tamara L.
%A Malik, Jitendra
%T Shape Matching and Object Recognition using Low Distortion Correspondences
%I EECS Department, University of California, Berkeley
%D 2004
%@ UCB/CSD-04-1366
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2004/6200.html
%F Berg:CSD-04-1366