Algorithm for segmentation based on an improved three-dimensional Otsu's thresholding

Considering the misclassification of conventional regional division based on gray level-average gray level-median gray level three-dimensional (3D) histogram, the result of 3D Otsu's method is not accurate enough. Its computing complexity rapidly increases due to the increment of another eigenvalue. Thus a fast iterative algorithm of the improved 3D Otsu's method is proposed. Based on the gray level-gradient level-median gray level 3D histogram, the improved regional division and fast iterative formulas of 3D Otsu's method are derived. The segmentation results, threshold and running time are given in the experimental results and analysis. A comparison is made with the fast iterative algorithms of the traditional 3D Otsu's method. The experimental results show that the proposed algorithm is more accurate for segmentation, and the detail features are more clearly. The running time reduces approximately 80% because there is no need to ransack the entire solution space while searching the best threshold.