Fabric defect detection in frequency domain using fourier analysis

An overwhelming majority of image processing based defect detection approaches rely on machine learning methods to train a model for comparison of test examples. This requires a training phase for each item to be learned and costly computations to tune model parameters. The fabric of textile always has repeating patterns that lends itself to automating the training phase by extracting a template. We avoid computationally costly machine learning methods by simple comparison of the fabric template with test examples in the frequency domain. In this thesis we show that it is possible to do online and fully automated defect detection of textile products in real time. We propose a method that leverages Fourier transform of textile images and present results on a data set that is collected in the scope of this research.