基于神经网络及多层次信息融合的手写体数字识别

以信息融合技术为基础，提出了一种新的基于神经网络及多层次信息融合的手写体数字识别方法。该方法通过提取字符图像不同机制的4个互补特征，组合形成6个融合特征，利用优化的BP神经网络算法，对多融合特征进行识别分类，然后用神经网络对6个识别结果进行融合决策．实验结果表明，新的融合识别方法能有效提高识别率，并具有较高的系统可靠性。

Handwritten Digit Recognition Based on Neural Networks and Multi-structure Information Fusion

According to the concept of information fusion technique, a new handwritten digit recognition method based on neural networks and multi- structure information fusion is given. The four different compensated features are extracted from the char image and six fusion features are obtained from fusing these features. An optimized BP neural networks was applied to classify the pattern with the multi-features at first, then the neural networks is used to fuse the results derived from the six classifiers. The experiment results shows that the new algorithm can increase recognition rate effectively, and the system has higher system reliability.