This paper presents a method for single image scale-up using sparse-representation modeling. The goal is to recover a high-resolution image from a low-resolution, noisy, and blurred version. Since the problem is ill-posed, a prior is needed to regularize it. The authors build upon a previously proposed algorithm by Yang et al. and assume a local sparse-land model for image patches as regularization. They introduce several modifications to improve performance, including a simplified algorithm with reduced computational complexity and a training approach for the dictionary-pair. The algorithm can operate without a training set by bootstrapping from the low-resolution image. The results are demonstrated on real images, showing visual and PSNR improvements. The single image scale-up problem is formulated as recovering a high-resolution image $ y_h $ from a low-resolution noisy version $ z_l $. The problem is to find $ \hat{y} \approx y_h $, given $ z_l $. Due to the Gaussian noise, maximum-likelihood estimation is achieved by minimizing $ \|SH\hat{y} - z_l\|_2 $. However, since $ SH $ is rectangular, it cannot be inverted stably. Existing methods use various priors to stabilize the inversion, such as Tikhonov regularization, Total-Variation, and sparsity of transform coefficients. The authors use the sparse-land model, which assumes that each image patch can be represented as a linear combination of a few atoms from a dictionary. This model is similar to previous work but differs in several important aspects. The paper is organized into sections describing the incorporation of the sparse-land model, algorithm implementation, experiments, and conclusions. The work assumes that the blur is applied before subsampling, and the scale-up process aims to reverse both steps. This generalizes the deblurring problem and can be considered a deblurring method without subsampling. Another line of work separates deblurring and up-sampling, assuming no blur or measuring performance based on the blurred high-resolution image. The authors' approach integrates both steps, aiming to achieve better results.This paper presents a method for single image scale-up using sparse-representation modeling. The goal is to recover a high-resolution image from a low-resolution, noisy, and blurred version. Since the problem is ill-posed, a prior is needed to regularize it. The authors build upon a previously proposed algorithm by Yang et al. and assume a local sparse-land model for image patches as regularization. They introduce several modifications to improve performance, including a simplified algorithm with reduced computational complexity and a training approach for the dictionary-pair. The algorithm can operate without a training set by bootstrapping from the low-resolution image. The results are demonstrated on real images, showing visual and PSNR improvements. The single image scale-up problem is formulated as recovering a high-resolution image $ y_h $ from a low-resolution noisy version $ z_l $. The problem is to find $ \hat{y} \approx y_h $, given $ z_l $. Due to the Gaussian noise, maximum-likelihood estimation is achieved by minimizing $ \|SH\hat{y} - z_l\|_2 $. However, since $ SH $ is rectangular, it cannot be inverted stably. Existing methods use various priors to stabilize the inversion, such as Tikhonov regularization, Total-Variation, and sparsity of transform coefficients. The authors use the sparse-land model, which assumes that each image patch can be represented as a linear combination of a few atoms from a dictionary. This model is similar to previous work but differs in several important aspects. The paper is organized into sections describing the incorporation of the sparse-land model, algorithm implementation, experiments, and conclusions. The work assumes that the blur is applied before subsampling, and the scale-up process aims to reverse both steps. This generalizes the deblurring problem and can be considered a deblurring method without subsampling. Another line of work separates deblurring and up-sampling, assuming no blur or measuring performance based on the blurred high-resolution image. The authors' approach integrates both steps, aiming to achieve better results.