摘要
通过分析文本特征和背景,提出一种基于最大梯度差的叠加文本定位算法。首先获得横向和竖向两个方向的梯度图像,然后设定一个窗口扫描整个图像,分别计算窗口内的最大梯度差,得到两个方向的最大梯度差矩阵,然后分别通过自适应阈值算法找出疑似文本像素,再将两个方向的判决结果取交集,消除部分复杂背景造成的误判。接着利用数学形态学运算和先验知识剔除伪文本区。最后利用改进的穿越线算法精确定位文本。实验表明,本算法不仅对横向文本具有较高的查全率和较低的虚警率,并且对竖向文本也有较好的定位效果。
This paper proposed an algorithm with max gradient difference by analyzing the text feature. It firstly calculated the gradient of two direct, vertical and horizontal. And then it got the max gradient different matrix by calculating the max gradient difference in a window. Then, it took an adaptive threshold algorithm to determine the text pixels, and calculated the intersec- tion of two results in order to eliminate the influence of part of the complex background. It conducted mathematical morphology operation and prior knowledge to eliminate the false text area. Finally, it used the improved across-line algorithm for precise locating of text. Experiments show that this algorithm not only has higher recall ratio of transverse text, and also has good effect for vertical text.
出处
《计算机应用研究》
CSCD
北大核心
2014年第10期3173-3176,共4页
Application Research of Computers
基金
国家"863"计划资助项目(2011AA010603
2011AA010605)
关键词
最大梯度差
叠加文本
文本定位
穿越线算法
max gradient difference
graphics text
text detection
across-line algorithm
