Fast Image Deblurring Algorithm Based on Normalized Sparsity Measure and Space-Frenquency Transformation

Blind restoration of blurry image is a challenging and significant problem. In this paper, we propose a deblurring algorithm which restores the latent image from a single blurry image. The method consists of two parts, kernel estimation and image restoration. To estimate the blur kernel, a cost function is constructed using a regularization term based on normalized sparsity measure and a fast optimization algorithm is employed to achieve the optimal solution based on space-frequency transformation. For image restoration, we construct the cost function through seeking the MAP estimation based on natural image gradient distribution, and solve it with a similar fast optimization algorithm. The experiment results with real natural images manifest that our method is able to obtain higher quality restored images with higher proceeding speed than other methods from current literatures.