应用MSER与DLBP的证件文本定位方法

ID photo text localization using MSER and DLBP

针对复杂背景文本定位问题,提出了一种应用MSER算法与DLBP纹理特征的文本定位方法.该方法由候选区域提取、文本行合并以及文本区域检测所组成.首先,使用MSER算法在预处理后的证件图像中进行多通道候选区域提取;接着,采用一种新设计的文本行合并算法对候选区域进行文本行合并;最后,利用DLBP纹理特征结合SVM分类检测从而确定最终的文本区域.实验结果表明,本文所设计的文本行合并算法结合DLBP纹理特征进行分类检测,能够得到较高的召回率和准确率.同时,对不同类型的证件都具有较好的兼容能力.

To resolve detection text line in complex background implementing a method combined with MSER and DLBP feature.The method consists of candidate region extraction,text line merging and text region detection.Firstly,MSER is used to extract candidate areas in multi-channel of preprocessed ID photo.Then a newly designed text line merging algorithm is utilized to merge the candidate areas.Finally,DLBP texture features and SVM are utilized to classify merged regions filtering out non-text region.Experiments show that the text line merging algorithm designed in this paper and using DLBP feature to classify the region can achieve high recall rate and accuracy,and it has good compatibility ability for different types of ID photos.