Surface defect detection on optical devices based on microscopic dark-field scattering imaging

Abstract

Methods of surface defect detection on optical devices are proposed in this paper. First, a series of microscopic dark-field scattering images were collected with a line-scan camera. Translation transformation between overlaps of adjacent microscopic dark-field scattering images resulted from the line-scan camera's imaging feature. An image mosaic algorithm based on scale invariance feature transform (SIFT) is proposed to stitch dark-field images collected by the line-scan camera. SIFT feature matching point-pairs were extracted from regions of interest in the adjacent microscopic dark-field scattering images. The best set of SIFT feature matching point-pairs was obtained via a parallel clustering algorithm. The transformation matrix of the two images was calculated by the best matching point-pair set, and then image stitching was completed through transformation matrix. Secondly, a sample threshold segmentation method was used to segment dark-field images that were previously stitched together because the image background was very dark. Finally, four different supervised learning classifiers are used to classify the defect represented by a six-dimensional feature vector by shape (point or line), and the performance of linear discriminant function (LDF) classifier is demonstrated to be the best. The experimental results showed that defects on optical devices could be detected efficiently by the proposed methods.