Haar Adaptive Taylor-ASSCA-DCNN: A Novel Fusion Model for Image Quality Enhancement

Abstract

In medical imaging, image fusion has a prominent exposure in extracting complementary information out of varying medical image modalities. The utilization of different medical image modality had imperatively improved treatment information. Each kind’ of modality contains specific data regarding subject being imaged. Various techniques are devised for solving the issue of fusion, but the major issue of these techniques is key features loss in fused image, which also leads to unwanted artefacts. This paper devises an Adaptive optimization driven deep model fusing for medical images to obtain the essential information for diagnosis and research purpose. Through our proposed fusion scheme based on Haar wavelet and Adaptive Taylor ASSCA Deep CNN we have developed fusion rules to amalgamate pairs of Magnetic Resonance Imaging i.e. MRI like T1, T2. Through experimental analysis our proposed method shown for preserving edge as well as component related information moreover tumour detection efficiency has also been increased. Here, as input, two MRI images have been considered. Then Haar wavelet is adapted on both MRI images for transformation of images in low as well as high frequency sub-groups. Then, the fusion is done with correlation-based weighted model. After fusion, produced output is imposed to final fusion, which is executed through Deep Convolution Neural Network (DCNN). The Deep CNN is trained here utilizing Adaptive Taylor Atom Search Sine Cosine Algorithm (Adaptive Taylor ASSCA). Here, the Adaptive Taylor ASSCA is obtained by integrating adaptive concept in Taylor ASSCA. The highest MI of 1.672532 have been attained using db2 wavelet for image pair 1, highest PSNR 42.20993dB using db 2 wavelet for image pair 5 and lowest RMSE 5.204896 using sym 2 wavelet for image pair 5, have been shown proposed Adaptive Taylor ASO + SCA-based Deep CNN.