Abstract
This paper presents a transfer domain strategy to tackle the limitations of low-resolution thermal sensors and generate higher-resolution images of reasonable quality. The proposed technique employs a CycleGAN architecture and uses a ResNet as an encoder in the generator along with an attention module and a novel loss function. The network is trained on a multi-resolution thermal image dataset acquired with three different thermal sensors. Results report better performance benchmarking results on the 2nd CVPR-PBVS-2021 thermal image super-resolution challenge than state-of-the-art methods. The code of this work is available online.
Highlights
Single image super-resolution (SISR) is a classical computer vision problem that tries to infer a high-resolution (HR) image from a single low-resolution (LR) input image
The approach that reaches the best result in peak signal-to-noise ratio (PSNR) metric uses the attention module; it achieves the third-best result in Structural Similarity Index Measure (SSIM)
The usage of just the first dataset shows a good performance; this means that this dataset has a large enough variability to train a network and that it is possible to perform a single thermal image super-resolution between two different domains using images acquired with different camera resolutions and without registration
Summary
Single image super-resolution (SISR) is a classical computer vision problem that tries to infer a high-resolution (HR) image from a single low-resolution (LR) input image. Thousands of HR images can be used for such a task; in the thermal image domain, most of the available datasets tend to have a poor resolution or do not present a high variability needed to generalize the training Due to this lack of thermal images, a novel dataset was proposed in [16] with three different resolutions (low, mid, and high) obtained with three different thermal cameras. Keeping in mind the limitation mentioned above of lack of large thermal image datasets, a novel CycleGAN architecture is proposed in the current work It is based on the usage of a novel loss function (SOBEL cycle loss) together with an attention module (AM) in the bottleneck of the generator.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
