An Adaptive and Fast Valley Emphasis Multilevel Otsu Thresholding Algorithm

The multilevel thresholding problem is a challenge task due to the fact that the computation is usually very time-consuming for obtaining the optimal multilevel thresholds. Though the state-of-the-art multilevel thresholding algorithms applied various meta-heuristic techniques or acceleration strategies, they still directly searched the optimal thresholds in the whole histogram only for the fixed thresholds number given by the user. Considering that the optimal thresholds usually locate at the valleys of the histogram, we propose an adaptive and fast valley emphasis multilevel Otsu thresholding algorithm (AFVEO). We constrain the searching space in locations of all valleys of the histogram, and it can greatly reduce the iterations required for computing the between-class variance due to the fact that the number of valleys is much fewer than the size of the histogram. And most important we can obtain the optimal multilevel thresholds for different thresholds number without fixed one given by the user. The experimental results indicate that the proposed method is more efficient than traditional Otsu method, recursive Otsu method, valley emphasis Otsu method and neighborhood valley emphasis Otsu method.