Thermal Image Super-Resolution Using Zero-Shot Super-Resolution Generative Adversarial Network (ZSSRGAN)

Thermal imaging has emerged as a vital tool in various fields, including surveillance, medical diagnostics, and industrial inspection, owing to its ability to capture temperature variations in the form of images. However, thermal cameras often have limited spatial resolution, compromising their effectiveness in many applications. Thermal image super-resolution is used in this research work to address this limitation, employing a Zero-Shot Super-Resolution Generative Adversarial Network (ZSSRGAN) model. The proposed framework combines the strengths of zero-shot super-resolution (ZSSR) and generative adversarial networks (GANs) to enhance the spatial resolution of thermal images. The zero-shot super-resolution model serves as the generator in the GAN architecture, converting low-resolution (LR) thermal images into high-resolution (HR) counterparts. Simultaneously, to create realistic and detail-rich super-resolved images, the discriminator network of the GAN is trained to discriminate between created HR images and actual HR thermal images. The results have been demonstrated on various datasets and compared the results using structural similarity (SSIM) and peak signal-to-noise ratio (PSNR) metrics. The suggested approach has shown improvement over current leading-edge models.