Novel Single Image deblurring Technique using Sparse representation and Discrete Radon Transform

In this paper, a novel single image deblurring technique based on sparse representation and radon transform is presented. The Sparse representation is used to make an initial estimation of the latent sharp image. Then, a set of directional filters are applied to the blurry and noisy image to reduce the noise while maintaining the blurry information on the orthogonal direction. After that an initial kernel estimation at different angles is performed using the initial latent image that was produced using sparse representation. Each estimated kernel is transformed to radon transform and the set of projections at different angles are stored. An inverse radon transform is applied to make a final kernel estimation. Finally, the Wiener deconvolution is performed to estimate the final latent sharp image. Experimental results showed the effectiveness of the proposed technique. The best obtained PSNR is 28.233 at 0.04 noise ratio, whereas the best obtained SSIM is 0.896769 at also 0.04 noise ...