Analysis and Experimental Research of Modifications of the Image Segmentation Method Using Graph Theory

We investigate the problem of image segmentation based on the color difference of regions. The aim of the research is to increase the processing speed of the image segmentation and improve the segmentation quality on textured images. We study the aspects of effective implementation of the image segmentation algorithm based on the minimum spanning tree graph, in particular, the use of different data structures for displaying segments. It is shown the dependence of the segmentation result on the color difference metrics. It is suggested the modification algorithm with the use of an array of singly linked list and with the sorting of the graph edges over linear time, which has resulted in 4 times speed gain. We propose the modification algorithm with the use of superpixelization, which avoids the re segmentation on the textured areas of the image achieved through the superpixel construction and its use as the graph nodes.