How accurately do different computer-based texture characterization methods predict material surface coarseness? A guideline for effective online inspection.

The growing industrialization has emphasized the need for high-performance computer-based inspection methods. Here, we investigated the performance of four major computer-based texture characterization methods in the prediction of visually perceived and actual surface coarseness of real materials. Gray level co-occurrence matrix (GLCM), distance-dependent edge frequency (DDEF), fractal dimension (FD), and histogram skewness (SK) were used as the methods. A novel collection of real materials consisting of 20 sandpapers with high, medium, and low coarseness levels was employed. The results revealed that at high coarseness level the most precise prediction of actual surface coarseness was made by GLCM and SK, while in the prediction of visual coarseness all the methods worked similarly effectively. Perfect correlations were observed between GLCM, FD, and SK at visual and also actual coarseness at medium coarseness level. At low coarseness level, SK and DDEF acceptably predicted visual and actual coarseness, respectively. The image resolution impact on performance of the computer-based methods was found to be substantial. Results of the research present a guideline for choosing the best computer-based method as a viable substitute for the human observer in online inspections of materials' texture.