手写数字识别的研究  

Research on handwritten digit character recognition

作　　者：杜梅[1,2,3,4] 赵怀慈[1,3,4] 

机构地区：[1]中国科学院沈阳自动化研究所光电信息研究室,辽宁沈阳110016 [2]中国科学院研究生院,北京100049 [3]中国科学院光电信息处理重点实验室,辽宁沈阳110016 [4]辽宁省图像理解与视觉计算重点实验室,辽宁沈阳110016

出　　处：《计算机工程与设计》2010年第15期3464-3467,共4页Computer Engineering and Design

摘　　要：为了提高手写数字识别的性能,研究了利用BP神经网络作为分类器在设计上的几个关键问题,给出每个关键环节的可行方案并进行有效总结。同时对脱机手写数字的图像预处理及特征提取部分的关键技术做了详细阐述。在此基础上给出分类器设计与训练的详细实验,实验结果表明,合理解决设计BP神经网络分类器的关键问题能够确保其对手写数字的高分类性能。In order to improve the performance of digit character recognition,some critical issues about design of BP neural network classfier are researched,and feasible solutions are given about every key issues,as well as effective summarization.In the mean time the two parts of image pre-processing and feature extraction about handwritten digit are described in detail.On this basis relevant experiment about classier designing and training is done.The experimental result shows that reasonable solutions to the key issues of the design of BP neural network classifier can ensure the outstanding performance of hand-written digit classification.

关 键 词：数字识别 BP神经网络 特征提取 预处理 分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]