Open AccessHWFusion: Holoentropy and SP‐Whale optimisation‐based fusion model for magnetic resonance imaging multimodal image fusion
Author(s) -
Venkatrao Patil Hanmant,
Damodar Shirbahadurkar Suresh
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.2017.0573
Subject(s) - image fusion , artificial intelligence , mean squared error , pattern recognition (psychology) , computer science , mutual information , wavelet , wavelet transform , whale , fusion , peak signal to noise ratio , particle swarm optimization , computer vision , mathematics , image (mathematics) , algorithm , statistics , linguistics , philosophy , fishery , biology
Image fusion is becoming a promising technique for obtaining a more informative image by combining various source images captured by multimodal imaging systems. The technique finds application in several fields, such as medical imaging, material analysis, satellite imaging, including defence and civilian sectors. This study presents a model, named holoentropy‐whale fusion (HWFusion), for the image fusion. Two different multimodal images from magnetic resonance imaging (T1, T1C, T2, FLAIR) are fed into the wavelet transform to convert the images into four subbands. The wavelet coefficients are then fused using a weighted coefficient that utilises two factors, entropy and whale fusion factor, which are calculated using holoentropy and the proposed SP‐Whale optimiser, respectively. SP‐Whale is an algorithm designed by modifying whale optimisation algorithm with self‐adaptive learning particle swarm optimisation and is used for the optimal selection of whale fusion factor. Inverse wavelet transform converts the fused wavelet coefficients obtained by the averaging of fusion factors into fused image. In a comparative analysis, the performance of HWFusion is compared with that of four existing techniques using, mutual information, peak signal‐to‐noise ratio, and root mean‐squared error (RMSE), where it could attain mutual information of 1.8015, RMSE of 1.1701, and peak signal‐to‐noise ratio of 40.6575.