Parameter estimation for LP regularized image deconvolution

Parameter estimation in Total Variation (TV) deblurring has been extensively studied in the literature during the last decade. However, few works have been done for parameter estimation in ℓ_p (0 <; p <; 1) regularized image deconvo-lution, although ℓ_p outperforms TV and ℓ₁ in natural image deblurring. In this paper, by utilizing the Bayesian framework, we propose an adaptive fast iteratively reweighted least squares algorithm for ℓ_p regularized image deconvolution, which automatically estimates the unknown image and regularization parameter. Experiments show that the proposed method yields nearly optimal results and outperforms the state-of-the-art methods.