This paper focuses on the problem Image Segmentation which aims at subdividing a given image into its constituent objects. Here an unsupervised method for color image segmentation is proposed where we first perform a Minimum Spanning Tree (MST) based “natural grouping” of the image pixels to find out the clusters of the pixels having RGB values within a certain range present in the image. Then the pixels nearest to the centers of those clusters are found out and marked as the seeds. They are then used for region growing based image segmentation purpose. After that a region merging based segmentation method having a suitable threshold is performed to eliminate the effect of over segmentation that may still persist after the region growing method. This proposed method is unsupervised as it does not require any prior informationabout the number of regions present in a given image. The experimental results show that the proposed method can find homogeneous regions present in a given image efficiently.
Keywords: Segmentation, Region Growing, Natural Grouping.