An accurate fuzzy measure-based detection method for various types of defects on strip steel surfaces

There are diverse types of defects with different forms, including scale, intensity, shape, and so on. It is a challenging task to detect all types of defects equally only with the same method. In this paper, a novel fuzzy measure-based method is presented to detect the defects on strip steel surfaces. By the statistical information of strip steel defect-free images, it is assumed that the background intensity of strip steel image obeys Gaussian distribution. Firstly, the histogram of a given test image is created, from which the model parameters of background are estimated. Then, a membership function is defined to estimate the extent to which each gray level belongs to defect, by which each pixel of test image obtains a membership value. So far, most of pixels can be detected as defect or background. Finally, combining the pixel connectivity, the maximum and sum of fuzzy connected regions are used to locate defects. Experimental results show that the proposed method achieves a high detection rate of 96.8%.