Improved Sparse Representation Based Super-Resolution

In sparse representation based super-resolution, high resolution image is estimated from a single low-resolution image using dictionary learning methods. The low-resolution image which has blur, noise and decimation, compared with original HR image. The missing high frequency details in low resolution image is learned from appropriate image database. This algorithm has ability to get back the lost information from the original image and their aim is to find the relationship between low-resolution image and high resolution image. The sparse vector used during dictionary learning, which consist of very less non-zero element is used to recover the patches. The phase congruency of Hilbert feature is used to sharpen the edges and to lower the ringing effects and jagged artifacts. The performance is increased by using K-SVD (singular value decomposition) dictionary learning algorithm, which update dictionary in each iteration and increases the quality of image. In our approach, which results in good quality HR image and having better PSNR values than the other similar SR methods