摘要
利用局部显著文字型区域描述结合连通分量分析思路,文章提出一种复杂背景下针对场景文本稳健的文本提取算法.该算法首先通过灰度的局部显著变化和显著多方向边缘强度定位图像的显著文本区域;然后依据区域灰度变化将原始图像进行分割,获得显著区域的连通分量包括文本分量和非文本分量;接着用一个级联分类器基于字符属性滤除部分非文本连通分量,最后统计文字包围盒的边缘方向分布信息,对每个候选连通分量提取8维的特征送入SVM分类器进行验证,最终定位出文本区域.实验结果表明,这种基于显著区域提取文本的算法有效地结合了文本的区域特性和文本连通体的笔画属性,在保证文本提取正确率的同时提高了算法速度,测试结果表现出较满意的性能.
Taking advantage of local text style salient region descriptor and ConnectedComponents(CCs)analysis,a robust scene text extraction in complex background approach is presented in this paper.Firstly,salient text region is located using local region's salient intensity variance and edge gradient in salient multiple orientation.Secondly,the image is decomposed into a list of CCs by local region's intensity variance,we acquire salient CCs containing text CCs and non-text CCs.Then most of non-text CCs are filtered out by cascade classifier and the remaining CCs are further verified by SVM.Thirdly,an 8-dimentional feature is extracted for each candidate CC.Finally we obtain the image containing text region only By SVM.The proposed algorithm is robust to combine text region's statistical characteristic and stroke feature of text CC.Experimental results demonstrate the method achieves impressive performance with respect to the accuracy,robustness and efficiency.
出处
《太原师范学院学报（自然科学版）》
2014年第4期33-39,50,共8页
Journal of Taiyuan Normal University:Natural Science Edition
关键词
显著文本区域
笔画特征
连通体分割
显著多方向
支持向量机
text salient region
stroke feature extraction
connected-Components analysis
salient multiple orientation
support vector machine
