Blind super-resolution (blind-SR) is an important task in the field of computer vision and has various applications in real-world. Blur kernel estimation is the main element of blind-SR along with the adaptive SR networks and a more accurately estimated kernel guarantees a better performance. Recently, generative adversarial networks (GANs), comparing recurrence patches across scales, have been the most successful unsupervised kernel estimation methods. However, they still involve several problems. x Their sharpness discrimination ability has been noted as being too weak, causing them to focus more on pattern shapes than sharpness. y In some cases, kernel correction processes were omitted; however, these are essential because the optimally generated kernel may be narrower than a point spread function (PSF) except when the PSF is ideal low-pass filter. z Previous studies also did not consider that GANs are affected by the thickness of edges as well as PSF. Thus, in this paper, 1) we propose a degradation and ranking comparison process designed to induce GAN models to became sensitive to image sharpness, and 2) propose a scale-free kernel correction technique using Gaussian kernel approximation including a thickness parameter. To improve the kernel accuracy further, we 3) propose a combination model of the proposed GAN and DIP(deep image prior) for more supervision, and designed a kernel correction network to propagate gradients through developed correction method. Several experiments demonstrate that our methods enhanced the l 2 error and the shape of the kernel significantly. In addition, by combining with ordinary blind-SR algorithms, the best reconstruction accuracy was achieved among unsupervised blur kernel estimation methods.

No abstract

The development of convolutional neural networks (CNN) has remarkably improved the current research on single image super-resolution (SISR). Several high-quality studies have been performed on reconstruction accuracy and perceptual quality, which are the two main issues in SISR. Nevertheless, numerous problems in SISR remain unsolved. SISR is inherently an ill-posed problem owing to insufficient information, and as the scale factor increases, the lack of information becomes even more pronounced. We have studied ways to solve the local characteristics of CNN to deal with additional useful information. A CNN uses a convolution layer designed based on local features, and repeatedly accumulates these features to expand a receptive field. We have explored network structures that can directly handle global information even at lower layers, which are not covered by the receptive field of a CNN. In this paper, we propose a non-local attention SISR network (NASR) that generates and utilizes the globally scattered similarity information of features. In addition, we propose a very deep architecture based on dense blocks that does not suffer from gradient vanishing without any normalization. Experimental results on standard benchmark datasets indicate the effectiveness of the proposed network, which exhibits state-of-the-art performance in terms of reconstruction accuracy and perceptual quality.

No abstract