摘要
化学分子结构图主要包括化学键和原子,主要研究如何提取静态图像中的端点原子信息。首先提取字符的网格特征和穿越特征组,利用BP神经网络进行训练、识别,接下来基于距离对端点字符组合识别。为了避免因个别字符识别错误而降低端点信息正确率,设计了一个端点原子参考表。将端点的识别结果与参考表进行比对,计算距离,选取与该结果距离最小的参考内容作为最终识别结果。从实验结果看,二维化学分子结构图端点信息的提取具有较高的准确率。
Chemical molecular structure images mainly include chemical bonds and atoms, and this paper focuses on the extraction of endpoint atom from static images. Gridding and traversing fea-tures of characters were extracted first, and BP neural network was used to train and identify characters. Secondly, the method of distance-based endpoint character identification was pro- posed. An endpoint atom reference table was designed to avoid character identification errors. The identified endpoints were compared with reference table, and the endpoint with the smallest distance was taken as the final identification result. Experiment results show that the method of identifying endpoints of two-dimension chemical molecular structure images has high accuracy.
出处
《淮海工学院学报(自然科学版)》
CAS
2013年第3期34-37,共4页
Journal of Huaihai Institute of Technology:Natural Sciences Edition
关键词
化学分子结构图
端点
神经网络
原子
chemical molecular structure images
endpoint
neural network
atom