目的 为了提高运动模糊图像盲复原清晰度，提出一种混合特性正则化约束的运动模糊盲复原算法。方法 首先利用基于保角单演相位一致性的边缘检测算法提取边缘细节，降低了噪声对边缘提取的影响。然后改进模糊核模型的平滑与保真正则项，在保证精确估计的同时，增强了模糊核的抗噪性能。最后改进梯度拟合策略，并加入保边正则项，使图像梯度更加符合重尾分布特性，且保证了边缘细节。结果 本文通过两组实验验证改进模型与所提算法的优越性。实验1以模拟运动模糊图像作为实验对象，通过对比分析5种组合步骤算法的复原效果，验证了本文改进模糊核模型与改进复原图像模型的鲁棒性较强。实验结果表明，本文改进模型复原图像的边缘细节更加清晰自然，评价指标明显提升。实验2以小型无人机真实运动模糊图像为实验对象，通过与传统算法进行对比，对比分析了所提算法的鲁棒性与实用性。实验结果表明，本文算法复原图像的标准差提升约11.4%，平均梯度提升约30.1%，信息熵提升约2.2%，且具有较好的主观视觉效果。结论 针对运动模糊图像盲复原，通过理论分析和实验验证，说明了本文改进模型的优越性，所提算法的复原效果较好。
Objective Motion blur blind restoration is the process of restoring a clear image without knowing the motion blur kernel function of image. How to solve the problem of resolving fuzzy kernel and the clear image accurately and efficiently is the key of blind restoration. The regularization constraint technique approximates the reconstructed image to the ideal image by properly adding regular items with prior knowledge in the model. The blind restoration problem can be solved quickly and effectively by this technique. Compared with single regularization method, the hybrid property regularization method can use known multiple prior conditions to impose multiple constraints on the model. While improving the accuracy of model solving, the number of iterations is reduced. Therefore, it is of great significance to study a multiple mixed feature regularization constraint method for motion blur blind restoration method. Aiming at the problems of poor anti-noise performance, incomplete fuzzy kernel smoothness constraint and edge blurring of the restored image appearing in the existing methods, a mixed feature regularization constraint method for motion blur blind restoration method is proposed. Method Firstly, in order to make the estimation of the fuzzy kernel model more accurate, the edge of the image is extracted to act on the fidelity term to estimate the fuzzy kernel. Edge extraction is easily disturbed by noise, leading to false edges or producing new noise, so as to reduce the estimation accuracy of the fuzzy kernel model. Therefore, the CMPM edge detection algorithm is used to detect the edge details of the blurred image to improve the anti-noise performance of the fuzzy kernel model. Then, in order to make the sparsity and smoothness of the fuzzy kernel, sparse regular items and improved smooth regular items are added in this paper. Because L0 norm smoothness regularization is complicated, L1 norm of fuzzy kernel is used to achieve the purpose of sparse constraint. Due to the insignificant effect of Tikhonov regularization term outlier value suppression, an improved multiple mixed regularization term is used to achieve the purpose of smoothing restraint while further suppressing outlier value. Finally, in order to make the restored image have a heavy-tailed character and further enhance the sharpness of the edge of the restored image, the super Laplacian prior term and the edge preserving regularization term are add in this paper. Due to the poor fitting effect of other prior distributions, the Laplacian prior is used to fit gradient distribution of the clear image to achieve the purpose of enriching the edge details. Since only the heavy-tailed property constrain is used to the restored image model, the edge resolution of the restored image is poor. Therefore, edge preserving regularization items are added to make the edge of the restored image closer to the sharpened edge of the clear image. Result In this paper, the advantages of the improved model and the proposed algorithm are verified by two groups of experiments. In first experiment, the simulated motion blur image is taken as the experimental object. By comparing and analyzing the restoration effects of the five combined steps, the robustness of the improved fuzzy kernel model and the improved restored image model are verified. The experimental results show that the edge details of the improved model restoration image in this paper are more clear and natural, and the evaluation index is obviously improved. In second experiment, the real motion blur image of a small UAV is used as the experimental object. The robustness and practicability of the proposed algorithm are compared and analyzed by comparing with the traditional algorithm. The experimental results show that the standard deviation of the restored image increases about 11.4%. The average gradient increases about 30.1%. The information entropy increases about 2.2%. And our method has better subjective visual effects. Conclusion Aiming at the blind restoration of motion blur image, the superiority of the improved model in this paper is demonstrated through theoretical analysis and experimental verification. The restoration of the proposed algorithm is better.