针对传统的正则化盲复原方法中图像和点扩散函数( PSF)的保真项和正则项分别采用同范数约束致使复原图像的质量下降, 估计出的PSF准确性降低的问题, 提出了多范数混合约束的正则化图像盲复原方法。首先, 复原图像的保真项和正则项分别采用L1范数和全变分(TV)范数来消除复原图像中的阶梯效应并较好地保护复原图像的边缘。其次, PSF的保真项和正则项分别采用L2范数和H1范数以降低估计PSF时参数调节的难度。最后, 通过分裂布雷格曼迭代方法对提出的模型进行最优化求解。在人造模糊图像和真实模糊图像上进行了实验, 结果表明: 提出的方法能够对运动、散焦等多种模糊类型的图像进行有效复原, 并准确地估计出相应的PSF。与近年来一些较好的模糊图像盲复原方法相比, 不仅在主观视觉效果上有较为明显的改进, 而且在客观的改善信噪比(ISNR)上也提高了0.36 dB到14.66 dB。

The traditional regularization blind restoration methods employ the same norm of the fidelity term and the regularization term for both the image and the Point Spread Function (PSF), which decreases the quality of restored image as well as the accuracy of estimated PSF. Therefore, this paper proposes a multi-norm hybrid constrained regularization method for image blind restoration. First, the L1 norm and the Total Variation (TV) norm are respectively adopted as the fidelity term and the regularization term for the image to eliminate the stair-casing effects and preserve the edges. Then, the L2 norm and the H1 norm are respectively adopted as the fidelity term and regularization term for the PSF to reduce the difficulty of the PSF estimation. Finally, the split Bregman iteration method is used to address the proposed model. Experiments are conducted on both synthetic and real-life degradations, and the results indicate that the proposed method can effectively restore a variety of artificial blurs such as motion blur, out-of-focus blur etc and can estimate the corresponding PSF accurately. Comparing to some other recent blind restoration methods, the proposed method can drastically improve the image quality in term of subjective visual and the Improved Signal-to-noise Ratio (ISNR) is improved from 0.36 dB to 14.66 dB.