Due to the severe noise and complex texture of blasted rock, selecting an appropriate binary segmentation algorithm is essential. This study evaluates nine common local thresholding methods for segmenting blasted rock images, with a focus on hyperparameter optimization of the highly adaptable Phansalkar algorithm. Experiments were designed to analyze the binarization effects of blasted rock images under different hyperparameters. The watershed algorithm and the findContours function from the OpenCV library were then used to perform feature parameter statistics, yielding segmentation results for various Phansalkar hyperparameters. Comparison with manual segmentation results indicates that the best segmentation is achieved with Phansalkar hyperparameters p = 1, q = 12, and k = 0.5. The corresponding correlation coefficient, root mean square error, and average relative error of the cumulative curve are 0.979%, 1.779%, and 5.53%, respectively, outperforming other hyperparameter settings.
Zhang et al. (Thu,) studied this question.