Intelligent Colorization for Thermal Infrared Image Based on CNN

Thermal infrared image coloring is an important issue to improve human perception in various fields, such as military, medical, security etc. This paper presents a novel algorithm to generate visual, i.e. perceptually realistic color(RGB) images from thermal infrared(TIR) images based on convolutional neural networks(CNN). The proposed approach uses an encoder-decoder architecture with skip connections. In the encoder-decoder framework, the input is passed through a series of layers that progressively downsample, until a bottleneck layer, at which point the process is reversed. There is an enormous low-level information shared between the input and output, and it would be desirable to shuttle this information directly across the net. To circumvent the bottleneck for information, we add the skip connections. We train our model on a KAIST-MS dataset and present illustrative qualitative and quantitative analysis of our results. Experimental results with a large set of original images provided the validity of the proposed approach.