Image segmentation and shape recognition by data-dependent systems.

摘要

Image processing techniques are widely used in industrial and military applications. Among them, image segmentation and shape recognition are two techniques crucial to automatic assembly, inspection, and automatic target recognition. Data Dependent Systems (DDS) methodology has been successfully applied to these two areas in this thesis. Thresholding is one of the most popular segmentation methods. Although many thresholding methods have been put forward by researchers, most of them require human intervention to select threshold values. A new approach of automatic multithresholding based on histogram modes is developed to threshold an image autonomously. Histogram characterization is achieved by autoregressive (AR) modeling via Levinson algorithm so that the histogram modes are defined and utilized for multi-thresholding. Object recognition and mensuration are important tasks in machine vision. To meet the challenge, shape and profile recognition techniques are developed and investigated. The shape and profile representations based on the DDS have been proven effective and robust in shape and profile recognition since they provide the inherent characteristics of the boundary geometry of an object. Therefore, the robust recognition and mensuration by machine vision can be accomplished.