Efficient Self-Adaptive Image Deblurring Based on Model Parameter Optimization

Natural images suffer from degradations in imaging system, and image blur is a major source of them. Most existing approaches aim to estimate a blur kernel via an alternating optimization method in multiscale space. However, in our practical project application, we need to deal with motion blurs come from moving conveyor belts. In this case, the degradation model and its orientation are known to us. In this paper, we propose a self-adaptive image deblurring method to deal with it. The model parameters are optimized by a heuristic algorithm, and the latent images are deblurred by a deconvolution technique based on ℓ
₁
-norm constraint. Simulation results show that our method not only acts on motion blur model, but also can deal with atmosphere turbulence model and defocus model, and the comparison results indicate that it outperforms others'. Furthermore, it is able to deal with motion blur in real scenes with high efficiency.