Single-image super-resolution using kernel recursive least squares

Online single-image super-resolution of an image has been obtained here. The high-resolution image is constructed from a dictionary of features that approximately spans the subspace of regression. This paper classifies the low-resolution image using the kernel k-means clustering algorithm and makes an extensive study using the approximate linear dependence kernel recursive least square and sliding window kernel recursive least squares for super-resolving the image from the existing low- and high-resolution images. The super-resolution using kernel recursive least square significantly provides an improvement up on the support vector regression solution, both in terms of speed, dictionary samples and also gives a better PSNR value.