Segmentation Using Adaptive Thresholding of the Image Histogram According to the Incremental Rates of the Segment Likelihood Functions

A novel algorithm for image segmentation based on adaptive thresholding of the global histogram of an image is proposed and applied to medical images from the medical database of the Second Department of Surgery of the University Hospital of Alexandroupolis, Greece. The threshold values are specified through an adaptive process that determines the optimum number of histogram regions and, at the same time, attempts to minimize the optimal likelihood value obtained from the specific partition of the histogram. The main peaks of the histogram are selected as seeds for the initial partition of the histogram. These seeds are subsequently grown by varying their upper and lower boundaries according to the incremental changes of likelihood values corresponding to the different intervals of the image histogram. The proposed method provides an alternative way of selecting the dominant peaks of the image histogram according to predefined constraints.