ROI DETECTION AND SEGMENTATION OF MEDICAL IMAGES USING OPTIMIZED THRESHOLDING AND CLUSTERING

Abstract

Image thresholding is one of the segmentation methods to isolate Regions of Interest from the images.    Maximum entropy is an image thresholding method that exploits entropy of the distribution in gray level of the image.  Clustering can also be applied as an image segmentation method to group pixels based on their intensity which in turn help to identify objects of interest from the image. Fuzzy clustering has been widely applied for recognition of patterns but has the shortcomings like ability to detect a data with same super spherical shapes, sensitive to initialization and convergence into local optima. Similarly, Ostu method is a simple and time saving technique but gives improper results on the noisy images and not able to detect ROIs when much intensity deviation among pixels is detected. But these limitations can be reduced when applied with optimization techniques like PSO and QPSO. This paper discusses Maximum Entropy, Fuzzy C-Means, Ostu, MEPSO, MEQPSO, FCMPSO, FCMQPSO, OstuPSO and OstuQPSO to segment  images and to find the objects of interests from  the images. The experimental section of this paper shows visual validation of these techniques and its performances.