基于角点类别特征和边缘幅值方向梯度直方图统计特征的复杂场景文字定位算法  

Text localization algorithm in complex scene based on corner-type feature and histogram of oriented gradients of edge magnitude statistical feature

作　　者：姜维[1] 卢朝阳[1] 李静[1] 刘晓佩[1] 

机构地区：[1]西安电子科技大学综合业务网国家重点实验室,西安710071

出　　处：《吉林大学学报（工学版）》2013年第1期250-255,共6页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(60872141);中央高校基本科研业务费专项基金项目(K50510010007);华为科技基金项目(HITC2011023)

摘　　要：针对复杂场景中纹理丰富的非文字区对文字定位算法的干扰,提出了基于光度不变量的角点类别特征和边缘幅值方向梯度直方图(Histogram of oriented gradients of edge magnitude,HOG-EM)统计特征两种新特征,并据此设计了一种两级多层复杂场景文字定位算法。首先获取边缘图像并提取根据HSL颜色空间特性划分的8层二值化图像,将其组成9层子图并做连通域分析提取文字候选区。然后提取文字候选区的角点类别特征和HOG-EM统计特征,将二者分别用于剔除非文字候选区和获取文字。实验表明:本文算法可以较为准确地剔除纹理丰富的非文字区,有效地降低复杂场景文字定位算法的虚警率,取得比较理想的准确率和召回率。The corner-type feature based on photometric invariants and the Histogram of Oriented Gradients of Edge Magnitude （HOG-EM） statistical feature are proposed to overcome the interference of the texture-rich non-text regions to the text localization algorithm. A two-stage multilayer text localization algorithm in complex scene is presented on the basis of the two novel features. In the proposed method, first, edge map is obtained and eight layers of binary maps in the Hue Saturation and Lightness （HSL） space are generated according to the characteristics of the HSL space. Then, nine layers of sub-maps are formed to gain text candidate blocks with multilayer connected component analysis. Finally, the two novel features mentioned above are extracted to remove the non-text block from the text candidate blocks and keep the text. Experiments indicate that the proposed scheme can efficiently remove non-text texture-rich regions, decrease false-alarm rate and obtain reasonable accuracy and recall rate.

关 键 词：信息处理技术 文字定位 角点类别 方向梯度直方图 光度不变量 

分 类 号：TN911.72[电子电信—通信与信息系统]