Defect Detection Using Unanimous Vote Among Mahalanobis Classifiers for Each Color Component

In manufacturing industries, product inspection is automated and the use of image data is increasingly being employed for defect detection. A manufacturing company in Japan produces an item and inspects the produced products using image data. Reducing the error rate is important in product inspection because poor inspection of products might lead to the delivery of defective products to consumers (consumer’s risk) and strict inspection increases production cost (producer’s risk). To reduce the error rate, we highlighted fault points using a two-dimensional moving range filter and discriminated defect production through a unanimous vote among Mahalanobis classifiers for each color component. For results, we achieved a lower error rate than the current system. This research is an empirical study of how to use image data in defect detection.