Segmentation and recognition of handwritten Pitman shorthand outlines using neural networks

Abstract

介绍了一种对英文Pitman速记发声字进行在线分割识别的新方法.该方法在预处理速记手写体的基础上,采用BP神经网络对发声字分割中可能出现的过分割进行检测和纠正,对音素记号/非音素记号和单个音素记号进行分类和识别,并实现了基于单个笔划识别结果的整体单词识别.通过对68个常用英文单词的测试,验证了该方法的平均识别正确率达到89.6%. Presents a novel approach for segmentation and recognition of on- line vocalized outlines of Pitman shorthand.The approach is to use a trained neural network for the segmentation of the vocalized outlines for the detection of over- segmentation; and to use another trained neural network for the recognition of Pitman shorthand consonant signs; while the word recognition was based on the estimation of the overall confidence on the stroke classification.Experimental results on a test set containing 68 most frequently used English words showed that on average,the approach can achieve an accuracy rate of 89.6%.