An experimental study of graph theory-based image segmentation techniques to improve graph cut algorithm

Abstract

The Graph cut algorithm of Graph theory has been successfully applied to various image segmentation problems. One of the most efficient algorithms is Normalized cut which is successfully implemented by many researchers for extraction of salient object from a natural scene. We have conducted a number of experiments on Normalized cut algorithm using Gaussian Filter, Median Filter, Lp1 norm based Gaussian filter and Lp1 norm based Median filter to enhance the efficiency of the Normalized Graph Cut Algorithms on individual color channels of an image and conducted different experiments to prove the efficiency of the new experimental models on the selected images from Corel and Berkley image segmentation data set. We have compared the results of the models with Otsu (Otsu thresholding algorithm) and C-means (C-means clustering algorithm), and to quantify the results we have used three validity indices.