Innovation of aggregate angularity characterization using gradient approach based upon the traditional and modified Sobel operation

Abstract

This study aims to employ the modified Sobel-Feldman operation to quantify aggregate angularity using the gradient approach, which has been used in previous studies. In detail, the modified Sobel-Feldman operation lies in expanding the conventional kernel size of the operation from 3×3 to 5×5, and up to 13×13 using the average gradient vectors of two neighboring pixels (2-pixel) instead of one single pixel (1-pixel) to calculate the angularity index. To achieve this goal, image processing was conducted to obtain the outlines of aggregate images. The gradient vectors of the surface pixels were obtained using the conventional and modified Sobel-Feldman operation. Then the angularity index (AI) of aggregates were calculated based on the change in gradient vectors of neighboring pixels with a rotational angle of 10°. The angularity index calculated using the conventional and modified Sobel-Feldman operations were compared. The sensitivity of the approach to image resolution was also discussed in this study. The profile images of both real aggregates and computer generated aggregate models were used for this investigation. The results show that the conventional Sobel-Feldman operation results have a higher Angularity Index than that of larger sized kernels. The AI values calculated from the 2-pixel method were more stable compared to those using the 1-pixel method. The modified Sobel-Feldman operation has lower sensitivity to image resolution than the conventional operation. Overall, the gradient approach using the 7×7 sized Sobel-Feldman operation with the 2-pixel method is the best way to calculate the angularity index. The findings of this study can potentially be adopted by commercial angularity index measurement apparatuses such as the aggregate imaging system (AIMS).