A super-resolution algorithm for thermal images

Abstract

Since the very first development of digital photography the research on images resolution enhancement, the detail level inside them,
has brought continuous improvements in image sensors. For the last decades a side-research about resolution enhancement has risen,
the super resolution of images. Super resolution algorithms are used to increase spatial resolution of low resolution images. There are
a lot of papers about super resolution applied on visible spectrum images; unfortunately that’s not true for thermal images, whose
resolution is not easily improvable through image sensors research. In fact, thermal images suffer from a lack of resolution due to
optical limitations and construction features. Higher resolution thermal images could improve analysis performances and expand
application possibilities. A super resolution algorithm has been developed in this paper. Starting from a set of low resolution images
with slight geometric and radiometric differences between each other, a single high resolution frame containing all their information
is reconstructed. The algorithm was then used to reconstruct different datasets, in order to test and calibrate its features and evaluate
achieved results. Reconstructed images are proven to have a better quality level than the low resolution ones. This improvement can
be seen through a visual comparison between them, but it’s also confirmed by the determination of statistical indexes.