A Structural Approach To Identify Defects In Textured Images

Abstract

From a structural point of view, a textured image is considered to be composed of skeleton and background primitives which occur repeatedly according to certain placement rule. Identification of defects in textured images is an important topic in computer vision. In this paper, an approach to defect detection is developed. A textured image is first thresholded using histogram analysis. Then it is mapped into a special data structure called the skeleton representation. Based on both location and length histograms from the data structure, several statistical measurements are defined. These measurements are used to identify and locate defects such as fluctuation, mean jump, and end influency in the textured image. The experimental results show satisfactory performance of this approach and its potential usefulness in industrial environments.