关键词: 图像盲复原, 相机异常值, 饱和图像, 超拉普拉斯先验, S型函数, 迭代加权最小二乘法

Abstract: Camera outliers seriously affect estimating the blur kernel, so the traditional image restoration algorithm is ineffective, serious loss of details, artifacts, this paper proposed a saturated blurred image blind restoration algorithm with removing camera outliers. First, the L1 regularization model is established according to the gray characteristics of the saturated image, and a hyper-Laplace prior is used to extract the salient edges of the image. Then aiming at the S function can not completely filter the saturated pixels in the edge, we propose a blur kernel auxiliary function which can effectively eliminate outliers by setting threshold. Finally, by analyzing the influence of outliers of blur kernel estimating ,we establish blind deconvolution model based on outlier-aware. Aiming at quadratic problem, the iterative weighted least squares method is used to obtain the restored image. Through experiments on several different saturated blurred images，the results show that average gray level gradient up to 12.689, image entropy up to 7.681，processing 255*255 images requires only 6.08s. It can effectively reduce the influence of camera outliers of estimating kernel, retain the integrity of clear details and significantly improve the speed of operation. Which is better than the most advanced image blind restoration algorithm.

Key words: Blind image restoration, Camera outliers, Saturated image, Hyper-Laplace prior, sigmoid function, the iterated weighted least squares