A novel histogram thresholding method for surface defect detection

One of the most important applications of machine vision in various industries is automated inspection. Performance of automated inspection depends directly on the algorithm used for threshold selection. Common methods of automatic thresholding are based on image histogram. In previous methods, the threshold selection has been realized by dividing the histogram into two classes. Also, possibility of misdiagnosis is high especially for the textures without defect. This paper proposes a new statistical algorithm for automatic theresholding which can be optimally applied in the presence of different types of surface defects. The optimum threshold is obtained in the proposed algorithm so that a maximum between-class and minimum within-class variances are provided. Proposed methods demonstrate a better performance compared to classic histogram-based algorithm particularly for the textures without any considerable defects.