基于布雷格曼迭代的稀疏正则化图像复原方法

为了实现模糊噪声图像的清晰化复原,提出了一种基于布雷格曼迭代的稀疏正则化约束的图像复原算法。首先,运用差分算子,得到图像中各个方向上的梯度信息;然后,利用提取的梯度信息,得到图像边缘各个方向上的权重;并结合稀疏性原理,针对复原图像,提出了一种权重的稀疏性正则化约束;最后,运用了一种布雷格曼迭代(Bregman Iteration,BI)策略对提出的方法进行最优化求解。实验结果表明,较近几年的一些具有代表性的图像复原方法相比,不仅主观的视觉效果得到了较为明显的改进,而且客观的信噪比增量也增加了0.3~2.5 dB。

A Bregman Iteration Sparsity Regularization Method for Image Restoration

In order to recover the blurred-noisy image, a Bregman-iteration based weighted sparsity regularization method for image restoration is proposed. First, using the difference operator, the gradient information of different directions in the image can be obtained. Second, making use of the gradient information, the weights for image edges in different directions can be obtained. Then, combining the sparsity theory, a weighted sparsity regularization constraint is proposed. Finally, a Bregman iteration (BI) approach is employed to restore the image. Experimental results indicate that the proposed method outperforms some representative image restoration methods, not only the subjective vision has the betterment obviously, but also the increase of the signal to noise ratio (ISNR) improves 2.2 dB.