Detecting fabric defects with computer vision and fuzzy rule generation Part I: Defect classification by image processing

A fabric defect detecting system that uses an advanced method involving computer vision and image analysis is capable of defect classification. Image pre-processing techniques that enhance raw images are applied before defect classification by a K-means algorithm and statistical method. These generate a Bayes classifier from which a decision surface is created for a classification procedure that can categorize defective or nondefective regions. Defect detection of the test fabric image is implemented by the decision surface from the training fabric image. The advantages of using the decision surface are a reduction in the training step and the ability to rapidly classify fabric. Experimental results confirm the reliable and reasonable classification ability of the proposed system.