Abstract: A general model for image restoration of basis function neural network is presented, a feed-forward neural network adopts a three-layer structure, where the hidden-layer neurons are activated by a group of orthogonal basis functions. To avoid lengthy BP-training of the iterative weights-updating, weights-direct-determination algorithm is proposed. Experimental results show that the algorithm not only can determine the weight by one step, but also gains higher accuracy.

Key words: image restoration, basis function neural network, orthogonal basis functions, weights-updating, weights-direct-determination

摘要： 给出了基函数神经网络图像复原的模型，该神经网络模型是由三层构成的前向神经网络，以一组正交基为隐层神经元的激励函数。为了避免反复迭代权值修正的冗长BP训练过程，提出了一种权值直接确定的算法。实验结果表明，该种权值直接确定算法不仅能一步确定权值而获得更快的运算速度，而且能达到更高的精度。

关键词: 图像复原, 基函数神经网络, 正交基, 权值修正, 权值直接确定