| Machine vision applications require image processing capable of reducing large quantities of data to easily interpretable and compact low level forms such as edges, regions, textures/features, and frequency decomposition. Traditionally, these low level image processing tasks have treated edge detection, image enhancement, and feature extraction as disjoint problems tuned to specific applications. However, experience has shown that the results of these image processing tasks are complementary to each other and lead to a more tractable, general purpose, and hence superior outcome. A much needed approach to such simultaneous and comprehensive processing of digital images based on the feature-model relationships derived solely from real images is presented in this dissertation.;Acquiring good quality images is a necessary prerequisite for the success of image processing and eventual machine vision applications. The preprocessing of digital images leaves much to be desired in terms of improvement in image quality. Hence, an effective analog video conditioning scheme based on digital image quality indicators and aimed at improving the image quality is presented and illustrated.;New schemes for robust estimation and outlier detection are tested on the real image scan lines and the results presented. The ;Important open research issues in the areas of image quality, model estimation, pattern recognition, and image representation are listed as suggestions for future work at the end. |
