Image-Alignment Based Matching for Irregular Contour Defects Detection

Abstract

Automatic defect inspection is attractive for high-quality workpiece manufacturing with irregular contours in order to achieve high accuracy and no contour defect. Thus, a novel image alignment-based feature matching algorithm framework is proposed in this paper. It can be used to solve the specified pixel-level defect detection and location problems for workpieces with irregular contours. A new forensic hash is first generated by extracting the scale, position, and main orientation information of feature points. Since the forensic hash is invariant to rotation, translation, and scaling, it is used for feature matching. A feature matching method based on a robust cascade estimator is proposed second to establish an accurate correspondence between the test image and reference image according to the obtained image hash and a parameter space voting mechanism. Third, the matched feature points are used to estimate the similar transformation parameters to achieve an accurate image alignment. Finally, image difference and morphological technique are used to locate the contour defect. Experimental results demonstrate that the proposed algorithm can effectively detect and locate small contour defects in irregular stamping workpieces.