Probabilistic models for 2D active shape recognition using Fourier descriptors and mutual information

Abstract

Shape recognition is essential for robots to perform tasks in both human and industrial environments. Many algorithms have been developed for shape recognition with varying results. However, few of the proposed methods actively look for additional information to improve the initial shape recognition results. We propose an initial system which performs shape recognition using the euclidean distances of Fourier descriptors. To improve upon these results we build multinomial and Gaussian probabilistic models using the extracted Fourier descriptors and show how actively looking for cues using mutual information can improve the overall results. These probabilistic models achieve excellent results while significantly improving on the initial system.