Image super resolution refers to increasing the physical dimensions of image to overcome resolution limitations originated from certain imaging sensors such as light sensitivity of sensor, environmental lightning conditions, physical size of sensor and also of each pixel w.r.t the sensor. With the help of sparse code and the dictionary it is possible to reconstruct good quality image with high resolution. Training images are used for obtaining the sample patches which are used for obtaining coefficients of the dictionary. Therefore the process of image super resolution is carried from sparse representation of low resolution image patch generating image of high resolution. Although method is effective, it is slow due to need of processing large number of image patches. Using LASSO (Least Absolute Shrinkage and Selection Operator) approximation which is one of the regression analysis method where the approximation of solution is performed. The implementation is done on MATLAB (CPU) and NVidia CUDA platform. Here NVidia GPU which increases the speed of patch processing thereby overall improvement time consumed to run the algorithm. The algorithm is evaluated using performance parameters SSIM, Running Time, RMSE, PSNR and Entropy. © 2018 IEEE.