WeAbDeepCNN: Weighted Average Model and ASSCA based Two Level Fusion Scheme For Multi-Focus Images

Abstract

Fusion of images is a strategy that merges various moderately focused images or non-focused images of a single scene to generate a fully focused, clear and sharp image. The goal of this research is to discover the focused regions and further combination of focused regions of different source images into solitary image. However, there exist several issues in image fusion that involves contrast reduction, block artifacts, and artificial edges. To solve this issue, a two level fusion scheme has been devised, which involves weighted average model along with Atom Search Sine Cosine algorithm-based Deep Convolutional Neural Network (ASSCA-based Deep CNN) and may be abbreviated as “WeAbDeepCNN” i.e. weighted average model and ASSCA based Deep CNN. In the study two images are fed to initial fusion module, which is performed using weighted average model. The fusion score are generated whose values are determined in an optimal manner. Thus, final fusion is performed using proposed ASSCA-based Deep CNN. The Deep CNN training is carried out with proposed ASSCA, which is devised by combining Sine Cosine Algorithm, abbreviated as SCA, as well as atom search optimization (ASO). The proposed ASSCA-based Deep CNN offers improved performance in contrast to current state of the art techniques with a highest value 1.52 of mutual information (MI), with a highest value of 32.55 dB of maximum Peak Signal to Noise Ratio i.e. PSNR as well as  value of 7.59 of Minimum Root Mean Square Error (RMSE).