液晶与显示, 2011, 26 (1): 116, 网络出版: 2011-02-21  

双X特征下自组织竞争网络的手写字符识别

Handwritten Character Recognition Based on Self-Organizing Network with Double-X Feature
作者单位
1 中国科学院 长春光学精密机械与物理研究所,吉林 长春 130033
2 中国科学院 研究生院,北京 100039
摘要
对图像预处理、字符特征提取和神经网络类型的选择、优化、训练及调整等进行研究,提出运用双X特征作为输入的自组织竞争网络对手写字符进行识别的方法。首先进行图像预处理,包括图像的灰度化、二值化、剪裁及缩放等。然后,提出字符的双X特征提取方法,并运用该方法提取各个字符的特征信息。最后,将字符特征数据送入自组织网络中进行学习,反复试验确定理想网络参数,使其可以自组织地将各字符模式相区分。实验结果表明:这种方法减少送入网络的数据量,降低了冗余信息对网络的干扰,使网络处理的复杂程度大大降低,训练后的网络误差小,网络的错判率约为12%,比传统方法有明显提高。
Abstract
By studying of image processing, character feature extraction and neural network including network type selection, network parameter optimization, training and adjustment, the handwritten character recognition method based on self-organizing network with double-X feature as input is proposed. First of all, the image is preprocessed including gray, binaryzation, tailoring and scaling. Second, the characters' double-X feature extraction method is utilized to extract feature information of various characters. Finally, the feature data of cha-racters is input into self-organizing network for learning and to determine the ideal network parameters after many tests, so that it can be self-organized to distinguish the character-mode. Experimental results show that this approach reduces the amount of data inputed to the network, and the interference of redundant information to network, and the complexity of network processing was considerably reduced.The network training error is small and the network misjudged rate is just about 12%.

仲崇亮, 丁亚林, 付金宝. 双X特征下自组织竞争网络的手写字符识别[J]. 液晶与显示, 2011, 26(1): 116. ZHONG Chong-liang, DING Ya-lin, FU Jin-bao. Handwritten Character Recognition Based on Self-Organizing Network with Double-X Feature[J]. Chinese Journal of Liquid Crystals and Displays, 2011, 26(1): 116.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!