Image Processing for Detecting Tile Defects Based on Adaptive Threshold

This paper addresses a new method of detecting tile defects. The tile image is changed to binary image by using a float threshold. Using linear regression of binary data gives a measure for classifying the defective. In order to detect a defective tile the defined measure is compared against a threshold which is estimated adaptively through a learning scheme using 4-folded cross validation algorithm. The choice of threshold Γ is determined through a simple training or parameter estimation stage depending on the type of tile texture. The results of applying this method to detect defects on both random and regular textured tiles are highly acceptable.