高阶混合正则化图像盲复原方法

提出了一种高阶混合正则化图像盲复原方法，用于实现模糊噪声图像的清晰化盲复原。根据自然图像边缘的稀疏特性，对图像的边缘细节成分进行了全变差(total variation TV)正则化约束，根据自然图像同性质平滑区域内像素值的变化规律，将一种高阶的类Tikhonov正则化约束运用于图像的平滑区域中，提出了一种新的高阶混合正则化模型。最后，提出一种多变量分裂布雷格曼(Multi-variable Split Bregman MSB)最优化迭代策略对提出的模型进行最优化求解。实验结果表明，提出的方法能够很好地保护图像的边缘细节，同时有效地消除图像平滑区域内的阶梯和假边缘瑕疵。与近几年的一些较好的图像盲复原方法相比，本文方法的信噪比增量(increase of the signal to noise ratio ISNR)增加了0.03~2.5 dB。

High-order hybrid regularization method for image blind restoration

A high-order hybrid regularization method for image blind restoration was proposed to restore blurry-noisy images blindly. Because of the sparse edges in a natural image, the Total Variation(TV) regularization restriction was applied to the edge texture component. According to the variation regulation of pixels in homogeneous smooth regions of the natural image, a high-order Tikhonov-like regularization restriction was applied to the smooth regions of the image, and a new model which combines the TV regularization restriction and the high-order Tikhonov-like regularization restriction was proposed. Finally, a multi-variable Split-Bregman (MSB) optimized iterative scheme was proposed to recover the image. A large number of experiments have been performed. The results prove that the proposed method is able to preserve the image edges while avoiding staircase effects and false edges in the smooth regions. The proposed method is compared with several recent image blind restoration methods, and results show that the Increment of Signal-to-noise Ratio(ISNR) has been improved between 0.03 dB and 2.5 dB.