MULTILEVEL THRESHOLDING METHOD BASED ON NORMALIZED CUT
Liying Zheng, KUIFENG LIU, Lei Yu · International Journal of Image and Graphics · 2009
Based on valleys of the histogram of an image as well as the normalized cut (Ncut) partitioning method, a novel multilevel thresholding technique with low computational complexity has been developed in this paper. A wavelet transform is first adopted to reduce the length of the original histogram, and to smoothen the histogram. All the valleys of the reduced histogram are then regarded as candida te thresholds based on which a weighted undirected graph is constructed. An Ncut criteria and a refinement procedure are finally adopted to determine the optimal thresholds. A comparison between the proposed multithresholding method and the methods based on GA, exhaustive search or iterative scheme has been done on benchmark images showing that our method outperforms other multilevel thresholding techniques in terms of computational complexity.