Date of Award
Santa Clara : Santa Clara University, 2023.
Computer Science and Engineering
Image super-resolution (ISR) is a prominent field in computer vision that aims to enhance the quality and resolution of low-resolution images. It finds widespread applications in various domains, including surveillance, satellite imagery, medical imaging, and media. The primary objective of ISR is to improve the visual quality and level of detail in low-resolution images. In this paper, we propose the Singular Value Decomposition and Transformer-based Image Super Resolution (SVDTIR) model, which combines both traditional and deep learning approaches to achieve image super-resolution. The SVDTIR model consists of three stages: image pre-processing, which employs the singular-value decomposition (SVD) algorithm to decompose the image into multiple separate images with different frequency components; the Transformer module, which enhances the resolution of the images; and image fusion, which combines the individual images into one final output. Through experimental evaluations, our model demonstrates good visual results compared to existing state-of-the-art models. It achieves a slightly higher peak signal-to-noise ratio (PSNR) and nearly the same structural similarity index measure (SSIM) when compared to these models. Furthermore, our model outperforms current online super-resolution models based on PSNR and SSIM metrics, showcasing its excellent performance. By proposing the SVDTIR model and showcasing its effectiveness, this research contributes to the advancement of image super-resolution techniques and opens up new possibilities for improving the visual quality of low-resolution images.
Qin, Andy; Zhang, Chen; and Cui, Timothy, "Singular Value Decomposition and Transformer-based Image Super Resolution" (2023). Computer Science and Engineering Senior Theses. 264.