This paper presents a method for image super-resolution using sparse representation. The approach is based on the principle of compressed sensing, where a low-resolution image is viewed as a downsampled version of a high-resolution image. The high-resolution image's patches are assumed to have a sparse representation with respect to an over-complete dictionary of signal-atoms. The sparse representation allows for the recovery of the high-resolution image from the low-resolution input, even when the input is a single low-resolution image. The method uses a dictionary of low-resolution patches, which are randomly sampled from training images. These patches are used to represent the low-resolution input, and the sparse representation is used to reconstruct the high-resolution image. The method is effective because it leverages the sparsity of the representation to regularize the otherwise ill-posed super-resolution problem. The algorithm is efficient and scalable, using linear programming to compute the sparse representation. It also allows for the adaptive selection of relevant patches from the dictionary, leading to superior performance compared to methods that use a fixed number of nearest neighbors. The method is tested on various images, including flowers, faces, and animal images. The results show that the method produces high-quality super-resolution images with sharp edges and clear textures. The algorithm is also compared with other methods, such as neighbor embedding and soft edge prior, and is found to produce better results in terms of image quality and reconstruction accuracy. The paper also discusses the preparation of dictionaries for the method, using random raw patches from training images. The use of derivative features is also explored, which helps in capturing the most relevant parts of the low-resolution signal. The method is shown to be effective with a small dictionary, which is significantly smaller than those required by other learning-based methods. The algorithm is summarized as a two-step process: first, finding the sparse representation of each local patch, and then using the reconstruction constraint to refine the entire image. The method is efficient and effective, producing high-quality super-resolution images with sharp edges and clear textures. The results demonstrate the effectiveness of sparsity as a prior for image super-resolution.This paper presents a method for image super-resolution using sparse representation. The approach is based on the principle of compressed sensing, where a low-resolution image is viewed as a downsampled version of a high-resolution image. The high-resolution image's patches are assumed to have a sparse representation with respect to an over-complete dictionary of signal-atoms. The sparse representation allows for the recovery of the high-resolution image from the low-resolution input, even when the input is a single low-resolution image. The method uses a dictionary of low-resolution patches, which are randomly sampled from training images. These patches are used to represent the low-resolution input, and the sparse representation is used to reconstruct the high-resolution image. The method is effective because it leverages the sparsity of the representation to regularize the otherwise ill-posed super-resolution problem. The algorithm is efficient and scalable, using linear programming to compute the sparse representation. It also allows for the adaptive selection of relevant patches from the dictionary, leading to superior performance compared to methods that use a fixed number of nearest neighbors. The method is tested on various images, including flowers, faces, and animal images. The results show that the method produces high-quality super-resolution images with sharp edges and clear textures. The algorithm is also compared with other methods, such as neighbor embedding and soft edge prior, and is found to produce better results in terms of image quality and reconstruction accuracy. The paper also discusses the preparation of dictionaries for the method, using random raw patches from training images. The use of derivative features is also explored, which helps in capturing the most relevant parts of the low-resolution signal. The method is shown to be effective with a small dictionary, which is significantly smaller than those required by other learning-based methods. The algorithm is summarized as a two-step process: first, finding the sparse representation of each local patch, and then using the reconstruction constraint to refine the entire image. The method is efficient and effective, producing high-quality super-resolution images with sharp edges and clear textures. The results demonstrate the effectiveness of sparsity as a prior for image super-resolution.