Optimized Convolutional Neural Network based Colour Image Fusion

Deep learning has been witnessing an unprecedented growth in various applications like image classification, image recognition, object recognition and so on. In this work, a novel multifocus fusion schematic is putforth using deep learning strategy for the fusion of more than two colour images. The activations of the convolutional neural network (CNN) are used to extract the prominent deep features of the source and these features are fused by the virtue of weighted averaging technique. Finally, the weighted average outputs of the activations of the source images are considered for the recovering the enhanced fused output the image. The fused image is found to be enhanced such that the entire image is free from motion blur and defocusing. Three popular deep learning architectures namely Alexnet, VGG16 and GoogLeNet are considered in this work. It is evident from the results presented in this work that, GoogLeNet based framework performs well when compared to Alexnet and VGG16.