基于MSER拟合椭圆的快速场景文本定位算法  

Fast scene-text localization algorithm based on MESR's fitting ellipse

作　　者：高士林[1] 吉立新[1] 李绍梅[1] 许喆[1] 

机构地区：[1]国家数字交换系统工程技术研究中心,河南郑州450002

出　　处：《计算机工程与设计》2015年第3期693-698,720,共7页Computer Engineering and Design

基　　金：国家863高技术研究发展计划基金项目(2011AA010603;2011AA010605)

摘　　要：针对图像中场景文本定位效率低的问题,提出一种基于最大稳定极值区的快速定位算法。快速提取图像中的最大稳定极值区;对提取的极值区进行椭圆拟合,构造MSER描述,利用拟合椭圆特征筛选极值区,滤除大部分伪字符区;运用以MSER为元素的投影合并算法,将字符区分割文本行并直接合并成单词,确定文本区。该算法通过拟合椭圆的特征选取实现伪字符区滤除,通过快速的投影合并算法,实现了高效的场景文本定位。实验结果表明,与其它算法相比,该算法能够达到较好的定位效果,在速度上具有优势。Text localization in natural scene images is an important prerequisite for many content-based image analysis tasks.A novel text localization algorithm was proposed.Firstly,a fast pruning algorithm was designed to extract maximally stable extremal regions（MSER）as basic letter candidates.These candidates were then filtered using the properties of fitting ellipse and the distribution properties of characters to exclude most non-text objects.Finally,a new extremal regions projection merging algorithm was designed to group character candidates into text candidates.Experimental results indicate that the proposed method has the advantage in speed and achieves relatively high precision and recall rates compared with the latest published algorithms.

关 键 词：场景文本 最大稳定极值区 拟合椭圆 伪字符区滤除 投影合并算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]