小波估计图像棱边分布的边缘保持规整化复原

Edge-preserving Regularized Image Restoration Based on Detail Distribution Estimation by Wavelet

为了从模糊图像中更准确地估计真实棱边分布,并将其用作边缘保持规整化复原的约束条件,提出小波多尺度分析的复原方法.该方法组合多个尺度的小波细节求相关,得到表征棱边分布的小波子带图像;在各向异性Markov随机场模型的基础上,选取满足边缘保持条件的势函数,将各方向小波子带滤波器代替梯度算子构造惩罚项,从而减少参数个数,降低训练复杂度;最后给出参数的自适应取值方法,并使用半二次规整化方法求解.实验结果证明,该方法复原的视觉效果优于梯度算子的方法,适合处理模糊尺度较大的图像.

The detail distribution of the real image estimated from a blurred image is important to the edge-preserving regularized image restoration. The method proposed in the paper combines wavelet details in several scales by seeking correlation operation. And the resultant image could reflect the edge distribution of the real image. Based on the Markov random field, proper potential function is chosen and the wavelet subband image is used to replace the gradient to construct the punitive function, so the number of regularization parameters and the training complexity are reduced. Finally, adaptive method is designed to assign the values to the parameters in the solution. Experimental results prove that the visual effect of the restoration result by this method is better than the common method using gradient, in particular better for processing the images in large degree of blurring.