Fabric defect detection based on golden image subtraction

To realise the universality and practicality of fabric defect detection in the textile industry, this paper proposes two approaches based on the Gabor filter and the golden image subtraction method. A method known as Gabor preprocessed golden image subtraction is first introduced, which filters a test fabric image by the real component of the Gabor filter with a 1 Hz centre frequency and a 90° angle. Golden image subtraction performs subtractions between the golden template and the filtered image to obtain a resultant image, and the segmentation threshold is determined by the direct threshold. The second method is Gabor preprocessed golden image subtraction based on a genetic algorithm, which can automatically select the parameter groups of the Gabor filter via the genetic algorithm. In addition, the paper also presents an extensive comparison between the proposed methods and wavelet preprocessed golden image subtraction. Meanwhile, the performances of the aforementioned three methods are tested in a real machine vision detection system to meet the actual demands of the textile industry. It can be concluded that Gabor preprocessed golden image subtraction provides the best detection results. The overall detection success rate is 95.62%, with 80 defect-free images and 80 defective images for fabric textures of common types.