Spatial Resolution Improvement of Thermal Infrared Images by Learning of Real Patch Pairs
Hideaki Osako, Katsuya Kondo, Kazu Mishiba · 2020
We propose the method to improve the spatial resolution of infrared images by exampled-based learning using patch pairs of real data of both low and high resolution images. When applying a super-resolution method to a color image, it is necessary to assume an image degradation model. In the conventional super-resolution technique for color images, down-sampling with bicubic interpolation is often used as a deterioration model. However, since infrared images have physical characteristics different from those of a color images, a color image degradation model is unsuitable for super resolution of infrared images. That is, it is difficult to represent a explicit function of degradation model. In the proposed method, the position of the raw infrared image is corrected, a dictionary of patch pairs at corresponding locations in the low-resolution image and the high-resolution image is created, and super-resolution based on the principle of sparse modeling is performed. From the experimental results, we show that our method using the degradation model based on actual data can restore the temperature distribution more accurately than the method using a degradation function model as down-sampling representation. Keywords : infrared images ,super-resolusion, thermal analysis