Suppression of Over-Segmentation in Watershed Segmentation C.Latha 1 , Dr. K.Perumal 2 1 Department of Computer Application, Madurai Kamaraj University, rocela16@gmail.com 2 Department of Computer Application, Madurai Kamaraj University, perumalmkucs@gmail.com ABSTRACT Image segmentation is one of the difficult tasks to mine information in the image process. To gratify ever- increasing necessity of image segmentation, various segmentation approaches have been tried over the past few years. A brain tumor is one of the most diagnosed illnesses which has affected many lives. Accurate detection of the ailment is the first step in the diagnostics. Experience indicates that it is extremely necessary to go into the segmentation of the MRI image before synchronizing the computer to go forward in the diagnosis. Watershed segmentation is a fast, robust and widely used tool in image processing as well as analysis although it suffers from over- segmentation. In this paper, attempts were made to improve the algorithm based on mathematical and morphological approach in the analysis and validation of brain tumor images. The results obtained through this work indicated improvements in the existing approach. KEYWORDS: Morphological operation, watershed ridge lines, Gaussian filter and SO-Segmentation. I. INTRODUCTION In magnetic resonance imaging (MRI) in the detection of brain tumor, segmentation is the foremost biomedical step in drawing information on the anatomical impairments as well as the presence of abnormal tissues. The segmentation of brain tumors is also being very useful for common modelling of pathological brains and also the development of pathological brain atlases. Despite various efforts and promising leads to the medical imaging, accurate and reproducible segmentation and characterization of abnormalities are considered as a challenge and difficult tasks for the reason that the shapes, areas and image intensities differ a lot in accordance with a variety of tumors. Some of them are known to deform the encompassing structures or may be associated with edema or necrosis which is likely to make changes on the depth of image around the tumor. Recent approaches depart from the need for large room for accelerated automation, applicability, and accuracy. Though imaging research connected with the detection of brain tumor by applying Magnetic Resonance images (MRI) is confronted with certain deficiencies it is still being used on the soft tissues of the human body Thus it is almost certain that image segmentation is essential in the diagnosis of the brain tumor.. Segmentation of the MRI image is a prerequisite for the computer to perform the accurate diagnosis. A variety of algorithms were developed in the past by employing different tools and procedures for the segmentation of MRI images. Furthermore, image segmentation may move towards perfection if the classification of all the image elements as pixels in an image group into special clusters which may indicate identical aspects. Segmentation involves partitioning of an image into a group of pixels that are homogeneous with respect to several criteria [7] . Diversified groups of image elements characterized by group-specific pixels shall not intersect with the nearest elements of heterogeneous image character. Such groups of image elements being referred as image segmentation are employed in the meaningful evaluation and interpretation of the image acquired. The grouping as driven by the pixel is an important component of an image analysis and pattern recognition system and thus considered as one of the most complex tasks in image processing, which resolve the feature of the final segmentation. 1.1. MAGNETIC RESONANCE IMAGING (MRI): An MRI Scanner [1] uses powerful magnets to polarize hydrogen nuclei i.e., the proton in the water molecule of human tissue, creates a detectable signal that is spatially encoded, resulting in images of the International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 10, October 2016 269 https://sites.google.com/site/ijcsis/ ISSN 1947-5500