Image Segmentation Using Extended Topological Active Nets Optimized Using Fuzzy Based Rules

Abstract

Segmenting the image and extracting the region of interest from the image is an newlineimportant part in image processing. Segmentation accuracy is very important for later newlinestage of image processing like image classification and recognition. newlineDeformable model based image segmentation is found to be very effective for images newlinewith complex coherent relationships. Topological active net and its extensions are newlinefound to perform efficiently among other deformable models. The extension to newlinetopological active net involves region and boundary based provisions. Despite their newlinegood performance, the existing Extended Topological Active Net (ETAN) newlineoptimization method involving limited search scope can lead to result inaccuracies. newlineThe segmentation inaccuracy is due to mesh optimization reaching local optima. Due newlineto local optima, the region of interest is not segmented with precise boundary and newlineedges are blurred. Towards this end many solutions have be proposed to prevent the newlinelocal minima problem in Extended Topological Active Net. newlineIn this research two different machine learning based improvement to ETAN is newlineproposed. newlineAn integrated fuzzy rule-based learning and objectiveness measurement-based newlineoptimization newlineDeep Learning based optimization newlineIn integrated fuzzy learning, a Fuzzy rule base is derived from training images for newlinewhich segmented result is available as ground truths. Fuzzy rule base aids in decision newlinefor placement of links at segmentation boundaries. The integrated fuzzy learning uses newlinetwo concepts of objectiveness based optimization and Fuzzy rule base. Mesh is placed newlineover the image and a objectiveness score is evaluated for each region to decide its newlinebelonging to foreground or background. All the background regions are removed. newlineOver the result of objectiveness based segmentation, Fuzzy logic based evaluation of newlinemesh portions on the boundary of the objects are done to decide whether they belong newlineto the object or not. The rule base for fuzzy logic is created based on training dataset. newlineDue to fuzzy evaluation,