Motion Deblurring Using Hybrid Fidelity Terms

摘要

Motion deblurring is a very hot and hard research topic currently. We propose a new deblurring method using hybrid fidelity terms and normalized Laplacian prior. The hybrid fidelity terms are introduced for using the gray and gradient information more effectively. The distribution of natural image's gradients is heavy-tailed. It is proved to be a good prior for image deblurring. The heavy-tailed distribution can be modeled as hyper Laplacian priors which is easy for implementation. The normalized hyper Laplacian prior can lead the energy decreasing while solving the deblurring equation. To reduce the complexity of solving the equation, a fast method called Split method is introduced. We use multiscale framework to enable the deblurring procedure from large scales to small scales until the blur kernel is got. The estimated kernel function and clear image are used as the initial value for the next scale. Using the estimated kernel, the final clear image can be got by total variation method. Experiments demonstrate the validity of the proposed method.