Object recognition using force data clustering and HMM based shape recognition

In this thesis the problem of detecting a known model object in a scene or database of images is addressed. We present two major components of a complete solution for this problem: a data clustering technique for image segmentation and feature extraction, and a shape recognition method. The presented novel data clustering method (Force) relies on the laws of electrostatic fields to find clusters of datapoints in a multiple-dimension space. Application of Force to image segmentation in gray level and color images is described in the thesis. We also show that Force can be successfully used for feature extraction from object images. We present a statistical shape matching method based on Hidden Markov Models (HMM) and then combine its recognition results with the recognition outcome of the Force based algorithm. We show improvement made when Force based features are added to the HMM based approach.