LNCS 4791 - A New Benchmark for Shape Correspondence Evaluation

A New Benchmark for Shape CorrespondenceEvaluationBrent C. Munsell, Pahal Dalal, and Song WangDepartment of Computer Science and EngineeringUniversity of South Carolina, Columbia, SC 29208, USA{munsell,dalalpk,songwang}@engr.sc.eduAbstract. This paper introduces a new benchmark study of evaluatinglandmark-based shape correspondence used for statistical shape analy-sis. Different from previous shape-correspondence evaluation methods,the proposed benchmark first generates a large set of synthetic shape in-stances by randomly sampling a specified ground-truth statistical shapemodel. We then run the test shape-correspondence algorithms on thesesynthetic shape instances to construct a new statistical shape model. Wefinally introduce a new measure to describe the difference between thisnewly constructed statistical shape model and the ground truth. Thisnew measure is then used to evaluate the performance of the test shape-correspondence algorithm. By introducing the ground-truth statisticalshape model, we believe the proposed benchmark allows for a more ob-jective evaluation of the shape correspondence than those that do notspecify any ground truth.1 IntroductionStatistical shape models have been applied to address many important appli-cations in medical image analysis, such as image segmentation for desirableanatomic structures [1,2] and accurately locating the subtle difference of thecorpus-callosum shapes between the schizophrenia patients and normal ...