中文摘要：

盲目图像复原的一个重要方法是检测出图像退化的点扩展函数的参数，然后根据估计的点扩展函数复原图像，但是对于高斯模糊的图像，很难检测出其点扩展函数，因此高斯模糊图像的盲复原一直是个棘手问题。本文提出一种新的算法：逆主元法，利用高斯点扩展函数的特性，在径向基神经网络的模型下，对图像进行盲目复原。实验表明，这种方法对轻度的高斯模糊图像，有很好的复原效果。

英文摘要：

In the area of blind image restoration, the deblurring of Gaussian blurred images is studied widely. After degraded by many factors, the image can be regarded as a Gaussian blurred image, hence it is important to deal with it without the PSF in prior. This paper presents a new method, inverse kernel method, which based on the model of radial basis function neural network. Experiments showed that, this method performed well for the light degradation and the counting of the algorithm was economic.