基于结构特征分类BP网络的手写数字识别

手写体数字识别有着重大的使用价值,用多层BP网络来识别手写体数字是手写体数字识别的一大进步,但是,用单纯的BP网络来识别也存在识别精度不高等的问题。将BP网络技术和数字本身的结构特征结合起来,提出了一种基于结构特征分类BP网络的手写体数字识别新方法。首先提取点、环等数字特征值,并根据一些特征进行分类;然后再运用BP神经网络识别,以提高网络的识别能力;最后,选取了500个人的0~9的手写体数字,运用以上算法进行BP神经网络识别,用3000个手写体数字作为训练样本,2000个其他的样本进行测试,网络收敛后,识别率达到96%以上,比原来有一定的提高。

Recognition of Handwritten Numerals with Grouped BP Net Based on Structural Features

Handwritten numeral recognition has important use value.It is a great advancement to use multi-layer BP network to identify numbers write by hand.But it has problems to use single BP network to identify.Combines BP network with the character of number structure,then a new method of grouped BP net based on structural features is proposed to classify handwritten numbers.Point and circle features cell are extracted and combined.Then,BP net is grouped based on some structural features.The system recognizes number by grouped BP neural network,therefore it obtains better effect.Finally,select the handwriting of 500 people from 0 to 9,using arithmetic to recognize(3000 swatches for training,2000 for testing).After net constringed,the distinguishing rate is over 96%.