A new deep-learning neural network for super-resolution up-scaling of thermal images
P. Więcek, Dominik Sankowski · Proceedings of the 2020 International Conference on Quantitative InfraRed Thermography · 2020
This paper presents a novel architecture of deep learning convolutional neural network for up-sampling of thermal images.The proposed solution is based on Kernel-Sharing Atrous Convolution (KSAC) filtering block.The developed system ensures the high accuracy of the up-sampling with scales up to 6 with much lower algorithm complexity compared to the reference methods widely used for visual image processing.The learning process uses the high resolution RGB visual images available in the DIV2K database.The examples of up-sampling of thermal images generated by IR cameras with160x120 and 32x24 sensors are presented.