摘要
考虑到字符的形状、大小、背景和对齐方式的多样性,提出基于笔画宽度构建多方向候选区域串的模型,用于检测任意定向和弯曲的场景文本。在提取图像最大稳定极值区域(MSER)的基础上,应用剪枝算法获取孤立的连通区域,应用笔画宽度变换(SWT)获得字符候选区域,使用丢失字符恢复的算法得到候选文本行,根据多向文本行的特征应用AdaBoost算法对文本行进行分类。仿真结果表明,该算法对任意笔画宽度、任意方向的文本均可以进行检测,取得了较好的效果。
Considering the diversity of the shape,size,background and alignment of characters,an algorithm based on the stroke width and possible character-string region was proposed to detect arbitrary orientation and bending scene text.On the basis of the maximum stable extremal region(MSER)of the extracted image,apruning method was proposed to obtain the isolated connected region according to the text,and the stroke width transform(SWT)was obtained according to the distance transformation.The connectivity rules were used to group the application,and the missing character recovery algorithm was used to get the candidate text line.The AdaBoost algorithm was used to classify the text lines according to the characteristics of multiorientation text lines.Simulation results show that the algorithm can detect the width of any stroke and any direction.
作者
方承志
黄梅玲
FANG Cheng-zhi;HUANG Mei-ling(School of Electronic Science and Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003, China)
出处
《计算机工程与设计》
北大核心
2018年第5期1377-1381,共5页
Computer Engineering and Design
基金
国家自然科学基金面上基金项目(61271334
61073115)
关键词
文本检测
最大稳定极值区域
笔画宽度变换
ADABOOST算法
丢失字符恢复算法
text detection
maximum stable extremal region(MSER)
stroke width transform(SWT)
AdaBoost algorithm
missing character recovery algorithm