Cluster analysis application to automatic flaw detection

The problem of automatic detection and categorization of flaws that arises in machine vision system development is discussed. A method of its solution is proposed. At the first stage, an autonomous process of system learning is performed where the learning samples of the analyzed flaw symptoms are split into clusters with medians. At the second stage, the real-time inspection process is performed where each incoming pattern is categorized according to the nearest median type. The paper notes that the comparative analysis against other techniques showed the effectiveness of the method proposed.