Comparative Study of Various Preprocessing Technique for CNN based Image Fusion

Image fusion is a technique of fusing every possible detail from the images of different modalities or different focuses into a single image so that a viewer can get the maximum quantity of information about the subject in one image. A tremendous effort has been made in this area with handcrafted techniques and later on utilizing deep learning methods. The authors here have endeavored to show the effect of preprocessing on the image fusion utilizing CNN. Three types of filters Gaussian, median and bilateral are employed discretely as a pre-processing step which shows median filter gives the best value of ISSIM in all three datasets (i.e. medical, infrared-visual, and color multi-focus dataset), additionally, Gaussian and bilateral filters gave the second-best values of NMI which denoted maximum quantity of mutual information is recuperated from different modalities.