REU/ CAREER: Deformable Shape Models for Image Understanding

点击翻译按钮获取中文摘要

The research plan is to develop shape representations for image and video ulderstanding. The approach is to encode shape and motion in terms of nonrigid deformations from prototypical or extremal views. This research builds on previous work in modal matching, an orthogonal shape decomposition for matching, describing, and comparing shapes despite sensor variations and nonrigid deformations. Shape and motion coefficients are expressed in terms of the eigenmodes of deformable shape models. The project extends this modal matching framework in four main directions; comparison metrics, prototype-based shape representations, including intensity information, and encoding motion and shape. Primary system evaluation is on synthetic data sets specifically designed to characterize algorithm limits with respect to noise and deformation. System performance is to be evaluated using a large database of images collected by a world wide web crawler. Towards this end, a web robot is being built. The educational plan addresses course development and teaching in multimedia computing, active engagement of undergraduates in research, and outreach for recruiting a greater number of students from traditionally underrepresented groups in computer science.