Application of vision technology in metal surface defect detection

The traditional metal surface defect detection needs to strengthen the surface features after soaking and air drying, and finally the defects are analyzed by manual visual inspection. Using machine vision technology, through the analysis of the image of the metal surface to be tested, the partial features of the defects are extracted, so as to realize the defect analysis of the metal surface. This paper compares Niblack algorithm and Sauvola algorithm of local threshold extraction, and introduces suppression factor on the basis of Sauvola algorithm, which improves the segmentation effect of Sauvola algorithm for areas with gentle gray change. At the same time, the integral image is used to optimize the algorithm, which further improves the time efficiency of the algorithm. After experimental comparison, the improved Sauvola algorithm can effectively detect scratches, stains and bumps on the metal surface.