基于有理数多项式先验模型的图像盲去模糊

图像盲去模糊问题是当今图像处理领域的热点问题之一.基于混合高斯先验模型的变分贝叶斯去模糊算法可以有效地复原模糊图像,成为一种重要的图像去模糊算法.虽然混合高斯先验模型可以很好地逼近自然图像的梯度分布,但是该模型在图像梯度值较大处往往会产生过拟合导致去模糊后的图像产生振铃效应,严重影响了图像可读性.利用有理数多项式先验模型代替混合高斯模型逼近自然图像的梯度分布,克服算法的上述缺点.有理数多项式函数的分母多项式强制函数在梯度值较大值时平滑,所以有效地避免了过拟合现象的发生,从而使得模糊核估计得更准确,减少振铃效应.实验结果表明了算法的可行性和有效性.

Blind image deblurring based on rational polynomial prior model

Nowadays blind image deblurring is one of the hot issues in the field of image processing. Variational Bayesian algorithm deblurring algorithm based on Gaussian mixture prior model can recover blurred image effectively, therefore it becomes a kind of important image deblurring algorithm. Although Gaussian mixture prior model can be well fitted to natural image gradient distribution, Gaussian mixture prior model causes overfitting at the big values of image gradient, so deblurring image undergo ringing artifact which can affect the image readability. Consequently, the method overcome above-mentioned algorithm disadvantage by using rational polynomial prior model to fit to natural image gradient distribution instead of Gaussian mixture prior model. The method can avoid overfitting effectively because denominator polynomial of rational polynomial function constrains smoothness of function at the big values of image gradient. The method makes estimation of blurring kernel more accurate, and reduces ringing artifact. The experimental results show the effectiveness and feasibility of the algorithm in this paper.