Automatic defects classification with p-median clustering technique

The problem of automatic defect recognition and classification for vision systems development is addressed. The main objectives of such systems are defect recognition and classification based on known features. The classification function is designed using cluster analysis. Two stages approach is proposed. On the first offline stage of classification a teaching process has been employed. On the second online stage inspection image is classified using its features comparison with the closest medians in real time. Comparative analysis with the state-of-the-art classification methods has demonstrated an efficiency of the proposed approach. Examples described here relate specifically to semiconductor industry but can be adopted to other manufacturing processes.