Multilevel image thresholding based on an extended within-class variance criterion

This paper addresses the issue of multilevel thresholding design for gray image segmentation. Most of the current multilevel image thresholding techniques require employing a criterion function to determine N-1 optimal thresholds for separating an image into N classes. In this paper, a new variance-based criterion function is proposed. Unlike the existing criterion functions, the proposed one is able to evaluate upper-bound and lower-bound thresholds for multiple classes individually. By doing so, it is possible to find 2N optimal thresholds for segmenting N classes. Moreover, an efficient multi-threshold searching is also proposed to speed up the threshold-decision process based on the proposed variance-based criterion function. Experimental results show that the proposed method not only performs well, but also succeeds to extract more details from background pixels.