Automated window size determination for texture defect detection

Texture defect detection methods are used for quality control purposes. Inspection systems fullfilling this task should be able to determine the necessary parameter sets automatically in case of a task change. One of these parameters is the size of the image window from which texture features are calculated. We present an approach for the calculation of an appropriate window size for statistical texture analysis methods. Our approach is based on the goal to choose the window size as small as possible to improve defect discrimination but to choose it also as big as necessary to achieve a sufficient texture representation within the window. The window size is calculated using a degree of feature deviation being defined. Experimental results are presented to show the relevance of the calculated window size.