Content area
Full Text
Abstract
Cracks in metals are commonly viewed as critical defects since cracks can lead to fatal accidents and cause great losses. Therefore, automatic crack defect detection based on image processing plays a vital role in improving the automation level of defect inspection in industry. This paper presents an automatic metal crack detection system based on a novel crack filter. To overcome the difficulties of image noises and nonuniform brightness, the proposed filter evaluates the image local brightness variation based on a Gaussian model. Moreover, multi-scale analysis is applied so mat the filter can enhance various sizes of cracks. As a result, the proposed crack filter is able to detect various sizes of metal cracks in the noise images with non-uniform brightness conditions. Experimental results show mat the proposed method can achieve crack detection accuracy up to 96.03% while maintains a low false ratio. It outperforms another algorithm based on Local Binary Pattern (LBP) by 26.62% with respect to the average detection accuracy, and by 6.23% with respect to the average false ratio.
Key Words: Crack detection, crack filter, X-ray image, multi-scale analysis, Gaussian model
1. Introduction
Metal crack detection is a very important process in modern industries since cracks can lead to fatal accidents and cause great losses. In most of the cases, metal cracks happen inside the metal and thus they cannot be seen with the naked eyes. Therefore, X-ray radiography, which is suitable for the detection of internal defects, is widely used in the manufacturing factories as a powerful and non-destructive technique. In a conventional inspection system, after capturing the X-ray images of the metal, the images are manually examined by an operator in order to determine if there are cracks. However, the radiologists can experience fatigue due to the long time working. Moreover, the inspection result can vary according to the skill and alertness of the operator. Therefore, to improve the inspection results and reduce the workload of the operator, the development of an automatic inspection system based on image processing is strongly required.
Crack detection based on image processing has been widely studied in the past years [1, 2, 3, and 4]. T. Yamaguchi et al [5] proposed an efficient and high-speed method for crack detection using percolation-based image processing....