An image thresholding method using a minimum weighted squared-distortion criterion

摘要

We present a new thresholding method for image segmentation in which thresholds are selected on the basis of a minimum weighted squared-distortion criterion. An iterative algorithm, which is named a supervised Lloyd algorithm, to satisfy the criterion is derived. In order to achieve reliable segmentation, we discuss how the weight coefficients of the criterion must be tuned by a supervisor. Two supervised approaches based on a probabilistic histogram model and typical images in processing images are treated in detail.