基于前向神经网络的与内容无关的笔迹鉴别

与内容无关的笔迹鉴别是属于图象处理和模式识别领域的一项课题，有着广泛的实用前景。本文提出了一种基于前向神经网络的与内容无关的笔迹鉴别的方法。文中讨论了提取笔迹灰度图象特征和用前向神经网络分类器进行两大问题。对笔迹灰度图提取了3大类18个灰度特征，面前向神经网络分类器由一种新的遗传算法同时优化设计其结构和权重矢量。通过对10人、每人6幅笔迹度图象用18个灰度特征进行鉴别,试验结果显示此方法设计的前向神经网络分类器收敛率高，比常用的最近邻分类器有更高的识别正确率。

A writer identification of text-Independent based on feed-forward neural betworks

The writer identification of text-independent is a subject in the area of image processing and pattern recognition,which has a wide range of potential applications. A feed-forward neural network method is proposed and applied to writer identification.The two problems that feature extraction from gray image and design of feed-forward neural network are discussed. Eighteen gray features of three sorts are extracted from a person's handwriting image. And a new genetic algorithm is proposed to optimize the structure and connection weight for feed-forward neural networks. Experiments have been conducted for writer identification with 18 gray features of per handwriting image from 10 persons. The result indicates the new genetic algorithm can optimize structure and connection weight of feed-forward neural networks together,converge to the global optimum quickly, hardly get stuck at local optimum, and increase the true rate efficiently.