Automatic Natural Image Segmentation by Using MarkerControlled Watershed Method and Region Merging Method

Abstract

Fully automatic image segmentation is a difficult task for natural images because of many variations ascontrast and complex background. Conventional segmentation methods require a considerable amount of interactive guidance by the user to attain satisfactory results. Moreover, the most subsequent tasks as object detection and image analyzing application highlydepend on the accurate and useful segmented result. Therefore, in this paper, an automatic image segmentation method for natural images is proposed. The proposed system includes three approaches: gradient computation with the modified LoG edge filter, marker-controlled watershed segmentation(MCWS) with automatically marker selection and region mergingapproach that is based on edge strength and homogeneous intensity. The system can not only efficiently reduce the significant over segmentation problem of watershed algorithm and but also produce the correct and meaningful segmented imagesIt purposes better performance of segmented images for image annotation, objects detection, image analyzing task and computer vision.