Image Segmentation Based on Minimal Spanning Tree and Cycles

A novel graph theoretic approach for image segmentation is presented in this paper. The image to be segmented is subjected to background elimination and then represented as an undirected weighted graph G. Each pixel is considered as one vertex of the graph and the edges are drawn based on the 8-connectivity of the pixels. The weights are assigned to the edges by using the absolute intensity difference between the adjacent pixels. The segmentation is achieved by effectively generating the Minimal Spanning Tree (MST) and thereby adding the non-spanning tree edges of the graph with selected threshold weights to form cycles satisfying certain criterion. Each cycle is treated as a region. The adjacent cycles recursively merge until the stopping condition reaches and obtains the optimal region based segments. This proposed method is able to locate almost proper region boundaries of clusters and is applicable to any image domain.