Colorization of Multispectral Image Fusion using Convolutional Neural Network approach

Abstract

The proposed technique  offers a significant advantage in enhancing multiband nighttime imagery for surveillance and navigation purposes., The multi-band image data set comprises visual  and infrared  motion sequences with various military and civilian surveillance scenarios which include people that are stationary, walking or running, Vehicles and buildings or other man-made structures. Colorization method led to provide superior discrimination, identification of objects (Lesions), faster reaction times and an increased scene understanding than monochrome fused image. The guided filtering approach is used to decompose the source images hence they are divided into two parts: approximation part and detail content part further the weighted-averaging method is used to fuse the approximation part. The multi-layer features are extracted from the detail content part using the VGG-19 network. Finally, the approximation part and detail content part will be combined to reconstruct the fused image. The proposed approach has offers better outcomes equated to prevailing state-of-the-art techniques in terms of quantitative and qualitative parameters. In future, propose technique will help Battlefield monitoring, Defence for situation awareness, Surveillance, Target tracking and Person authentication.