ROBUST COMPUTER VISION

Computer image recognition systems must estimate accurate model parameters despite small-scale noise in the data and occasional large- scale measurement errors. This monograph presents a framework for a maximum likelihood approach to solving parameter estimation problems in computer vision applications involving similarity, then applies the framework to experiments on color based retrieval of images and objects, robust texture analysis, shape based retrieval, stereo matching, motion tracking, and facial expression recognition. The authors (Leiden University) survey literature on robust techniques in each particular field, discuss the results of comparative experiments, and conclude each chapter with observations.