Defect Detection on Imaging Surfaces by Using Multi-Scale Image Analysis

Abstract

The proposed method for automatic defect detection is based on a time-effective algorithm of multi-scale and structure-adaptive analysis of pictures obtained from imaging surfaces. Application of the concept of image multi-scale relevance function in the framework of visual attention mechanism provides a quick and reliable location of regions of attention with potential defect indications considered as objects of interest on images. The relevance function is a local image operator that has local maxima at centers of the objects of interest or their regular parts, which are termed primitive patterns. A detailed structure-adaptive image analysis is performed within the regions of attention in order to make the final decision on defect presence in a current focus-of-attention point. The method requires a simple parameter learning procedure applied to sample images of imaging surfaces prior the automatic inspection. The testing results indicate on the superiority of the relevance function approach to location of defect indications in images over the known computer vision methods for defect detection.