Image thresholding by histogram segmentation using cluster organization avoiding local minima

lmage thresholding is usually used to distinguish the foreground from the background. The image thresholding problem is closely related to the clustering problem, which segments the image into several classes. This paper proposes a novel method of image thresholding using the optimal histogram segmentation by the cluster organization based on the similarity between adjaccnt clusters. The similarity measure proposed in this paper is based on the mean distance of the clusters to be marged and the compactness ofthe new merged cluster. Since this method is not based on the minimization ofa function, the problem ofselecting the threshold at the local minima is avoided. This approach overcomes the local minima that affect most of the conventional methods by maximizing the interclass variance and minimizing the intraclass variance. It is sraightforward to extend the method to multi-level thresholding problem by stopping the grouping as the expected segment number is achieved. Results obtained from automatic thresholding ofthe exp€rimental images are showing the validity ofthe method. Keyword Thresholding, Clustering, lnterclass variance, Intraclass variance l . In t roduc t ion Image segmenta t ion is very essent ia l to process ing and pa t te rn recogn i t ion . I t leads r igh qua l i t y o f the f ina l resu l t o f ana lys is . segmenta t ion is a p rocess o f d iv id ing an image in to image d i f fe ren t reg ions . One o f the spec ia l k inds o f to the segmenta t ion is th resho ld ing , wh ich a t tempts to Image c lass i fy image p ixe ls in to one o f the two ca tegor ies