中文摘要：

传统铭牌字符识别主要通过计算铭牌字符图片的灰度平均值来判定,由于铭牌字符具有笔画方向特征、轮廓特征,同时字符之间存在着一定的排列特征。提出融合字符及字符排列特征的铭牌识别方法,首先对铭牌字符图片分别进行横、竖、撇三个方向上的小波变换,求出三个方向上的小波平均能量,再提取出字符的边缘方向直方图,以小波平均能量和边缘方向直方图构成特征向量,用支持向量机分类器训练并构建候选字符识别模型,得到候选字符,然后利用铭牌字符排列特点和铭牌的样本数据训练构建N阶马尔科夫字符排列模型,借助于模型对候选字符进行约束获得铭牌识别结果,最后对电力设备铭牌进行识别实验。结果表明,提出的方法表现了很好的效果,比OCR软件识别的准确率提高了12.6%。

英文摘要：

The nameplate recognition method based on fusion characters and character arrangement features is proposed in this paper. At first, the wavelet transformation for nameplate character images is carried out in three directions of horizontal, ver- tical, and leftfalling and rightfalling strokes. Then the average wavelet energy in the three directions can be determined, and the edge direction histograms of the characters are extracted. The feature vectors are composed by the average wavelet energy and edge direction histogram. The candidate character recognition model is trained and constructed by support vector machine （SVM） to obtain the candidate character. An N-order Markov character arrangement model is built by using arrangement fea- tures of nameplate characters and sample data training for nameplates. The candidate characters are constrained with the help of model to obtain the nameplate recognition result. The results of experiments on power equipment nameplates show that the accu- racy rate of the proposed method is increased by 12.6% in comparison with that of optical character recognition （OCR） software.