Abstract: Locating the telephone number region is a very important technology in telephone number recognition of the fax images. After realizing a relative precise page analysis on the fax images by adopting the Connected Component Analysis (CCA) to form comparatively whole word regions, the features of horizontal traversing times and spatial distribution were abstracted to form feature vector of fifty-one dimensions. The multi-class Support Vector Machine (SVM) based on normal decision tree was introduced to achieve the key words location. The experimental results show that the method can realize the location quickly and effectively, and it is valuable in applications.

Key words: Connected Component Analysis (CCA), horizontal traversing times, spatial distribution feature, Support Vector Machine (SVM), key words location

摘要： 电话号码区域定位是传真图像电话号码识别中的关键技术之一。首先采用连通域分析对传真图像实现较为精确的版面分析，形成比较完整的单词连通域，提取单词连通域的水平穿越次数和空间分布特征，形成51维的特征向量。采用基于正态决策树的多分类支持向量机(SVM)，来完成对传真图像电话号码区域关键词的定位。实验结果表明，算法能够快速有效地完成关键词的定位，具有较强的实用价值。

关键词: 连通域分析, 水平穿越次数, 空间分布特征, 支持向量机, 关键词定位