International Journal of Software Engineering and Its Applications Vol. 10, No. 12 (2016), pp. 217-226 http://dx.doi.org/10.14257/ijseia.2016.10.12.18 ISSN: 1738-9984 IJSEIA Copyright ⓒ 2016 SERSC Super-Resolution Image Reconstruction with Improved Sparse Representation Muhammad Sameer Sheikh and Qunsheng Cao Nanjing University of Aeronautics and Astronautics (NUAA), Nanjing, China. sameer@nuaa.edu.cn Abstract In this paper, we present a new approach to reconstruct a high resolution (HR) image from a low resolution (LR) input image based on a two dimensional (2D) sparse method. The new method consists of three phases. Firstly, the nonlinear feature of the input LR image is divided into the linear subspace, and then LR-HR dictionaries are learned to reduce the blurred artifacts of the image. Secondly, 2D sparse representation and self- similarity are developed to strengthen and enhance the image structure. Finally, the final HR image is achieved by reconstruction of all HR patches. Simulation results demonstrated that our proposed method achieved superior results on real images, and shows various improvements in terms of PSNR and SSIM values as compared with some other competent methods. Keywords: image super-resolution, image enhancement, sparse representation, visual resolution 1. Introduction Super-resolution (SR) image is considered as a major area of research in digital image processing, and its offer to overcoming the limitation of low resolution images. Many applications have been benefited by this approach such as, biomedical imaging, and astronomical imaging model etc. The aim of super-resolution (SR) image is to reconstruct a visually pleasant high resolution (HR) image from one or more low resolution (LR) input images. In recent year, several SR imaging methods have been proposed, which can be categorized as an interpolation-based method, learning-based and reconstruction-based methods respectively. These techniques provide robust improvement to overcome the low visual resolution in SR imaging [1]. Interpolation-based SR methods are divided into bicubic interpolation and bilinear method respectively. These approach are termed as a simple and fast with less complexity, but they often generate visually displeasing image with sharp edges. The adaptive kernel method is used to evaluate the unidentified pixels present in an HR image grid [2-3], this method are limited to the real-time applications and mostly can generate blurred artifacts at reconstruction part. By using the externally trained datasets, the learning based method is used to reconstruct the HR image by evaluating and mapped between the pairs of LR-HR patches, this method produces blur artifacts and leads to be an unsatisfactory reconstruction because of relying on external training datasets [4]. Reconstruction-based method usually assume the prior knowledge of LR image, which is the combination of several degrading factors, such as blur, noise and down sampling operators. Consequently, many information is missing in a LR image, so one LR image is equivalent to many HR images, this lead to be an ill-posed problem. Recently, different algorithm have been proposed to inverse this problem, by utilizing the prior knowledge, such as, redundancy prior which is used in reconstruction-based method. However, this method can generate visually pleasant image with sharp edges and suppress artifacts [5- 6]. Super-resolution image based on 2D sparse has been proposed, this technique is the