An Automatic Thresholding for Crack Segmentation Based on Convex Residual

Abstract

Automatic thresholding segmentation has been widely used in machine vision detection. From no crack defect images to crack defect images on the beam surface of straddle-type monorail, the proportion of crack is zero or small comparing to the background, so the histogram distribution is unimodal or close to unimodal. The Otsu method can be successful if the histogram is bimodal or multimodal, but it provides poor results for unimodal distribution. In this paper, a segmentation approach based on convex residual is proposed, according to the convexity and concavity of histogram of detected image, the gray value which is the maximal value of convex residual joining with between-class variance is the threshold, experimental results show that the segmentation performance is superior to the Otsu method and the valley-emphasis method.