Open AccessFast and Automatically Adjustable GRBF Kernel Based Fuzzy C‐Means for Cluster‐wise Coloured Feature Extraction and Segmentation of MR Images
Author(s) -
gmeikapam Kishorjit,
Kumar Wahengbam Kanan,
Singh Aheibam Dinamani
Publication year - 2018
Publication title -
iet image processing
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.401
H-Index - 45
eISSN - 1751-9667
pISSN - 1751-9659
DOI - 10.1049/iet-ipr.2017.1102
Subject(s) - artificial intelligence , pattern recognition (psychology) , image segmentation , segmentation , scale space , cluster analysis , mathematics , computer science , kernel (algebra) , scale space segmentation , gaussian function , euclidean distance , computer vision , image (mathematics) , gaussian , image processing , combinatorics , physics , quantum mechanics
Fuzzy C‐means algorithm is a popular image segmentation algorithm and many researchers in the past have introduced several improved versions of it. However, they still lacked robustness for segmenting key regions of magnetic resonance (MR) human brain transversal images such as white matter, grey matter, and cerebro spinal fluid with an almost similar contouring of the edges. This study highlights a robust algorithm that is effective in four ways: (i) a distinct contrast between different regions of the human brain, (ii) reduce noise as a result of the contrast stretching procedure, (iii) efficient analysis using both grey scale segmented image and its colour segmented version, and (iv) striking a balance between time consumption and image segmentation results. These objectives have been achieved in two phases: first, MR image segmentation using fast and automatically adjustable Gaussian radial basis function kernel based fuzzy C‐means (FAAGKFCM) algorithm which utilises a new objective function that incorporates a similarity measure having both spatial and grey level measure factor ( S ij ), an adaptive factor ( w i ) to remove discrepancy of grey scale image, and using a Gaussian radial basis function (GRBF) kernel‐based distance metric in place of the traditional Euclidean distance metric. Second, converting FAAGKFCM segmented image into CIE L * a * b * colour space and successively performing cluster‐wise feature extraction using hard k‐means clustering.