摘要
讨论一种基于HMM(隐马尔可夫模型)的英文印刷体识别方法。先将整篇文本图像切分成字母级别,提取出字母轮廓的8方向特征,之后把特征向量进行矢量量化并送入HMM训练识别。根据切分中出现的错误特点,对矢量量化过程和训练算法提出一些改进方法,提高识别率。
Discusses a printing English words recognize method based on Hidden Markov Model.The entire text image is segmented into letters,and extracts each letter＇s 8-direction features,then adopts a vector quantization method and sends the feature vectors to HMM for training or recognition.According to segmentation,proposes some improvements in VQ method and training algorithm,so as to improve the recognition rate consequently.
出处
《现代计算机》
2011年第8期14-16,共3页
Modern Computer
关键词
特征提取
矢量量化
隐马尔可夫模型
Feature Extraction
Vector Quantization
Hidden Markov Model
