Single Image Super-Resolution via CNN Architectures and TV-TV Minimization
This work addresses the inconsistency issue in super-resolution for applications such as medical imaging and consumer electronics, offering an incremental improvement over existing methods.
The paper tackles the problem of single image super-resolution by combining convolutional neural networks (CNNs) with classic reconstruction-based algorithms to enforce consistency between super-resolved and low-resolution images, resulting in superior image quality with better PSNR and SSIM scores on datasets like Set5, Set14, and BSD100 compared to state-of-the-art CNN methods.
Super-resolution (SR) is a technique that allows increasing the resolution of a given image. Having applications in many areas, from medical imaging to consumer electronics, several SR methods have been proposed. Currently, the best performing methods are based on convolutional neural networks (CNNs) and require extensive datasets for training. However, at test time, they fail to impose consistency between the super-resolved image and the given low-resolution image, a property that classic reconstruction-based algorithms naturally enforce in spite of having poorer performance. Motivated by this observation, we propose a new framework that joins both approaches and produces images with superior quality than any of the prior methods. Although our framework requires additional computation, our experiments on Set5, Set14, and BSD100 show that it systematically produces images with better peak signal to noise ratio (PSNR) and structural similarity (SSIM) than the current state-of-the-art CNN architectures for SR.