基于色彩聚类的自然场景文本检测  

Text detection from natural scene images based on color clustering

作　　者：吴慧[1,2] 赵于前[1,2,3] 李香花[4] 邹北骥[2] 邹润民[2] 

机构地区：[1]中南大学地球科学与信息物理学院,湖南长沙410083 [2]中南大学信息科学与工程学院,湖南长沙410083 [3]中南大学有色金属成矿预测教育部重点实验室,湖南长沙410083 [4]中南大学土木工程学院,湖南长沙410075

出　　处：《中南大学学报（自然科学版）》2015年第6期2098-2103,共6页Journal of Central South University:Science and Technology

基　　金：国家自然科学基金资助项目(61172184;61174210;61379107;61402539);新世纪优秀人才支持计划项目(NCET130603);高等学校博士学科点专项科研基金资助项目(20130162110016);湖南省科技基本建设项目(20131199);中南大学中央高校基本科研业务费专项资金资助项目(2015zzts052)~~

摘　　要：基于场景图像背景复杂,图像中的文本变化多样,提出一种基于色彩聚类的自然场景文本检测算法。其步骤为:首先,用K-均值聚类算法将彩色图像聚成5种色彩层,分析各色彩层中连通域外接矩形的几何特性,去掉离散的非文本连通域;然后,根据相邻连通域外接矩形的几何性质和空间位置关系,将离散的连通域连接形成候选文本块;最后,通过分析候选文本块的几何特性和边缘密度验证文本,得到最终文本检测结果。实验结果证明了本文算法的有效性和可行性。Considering that the natural scene images have complex background and the text in them is usually various, a text detection method based on color clustering was proposed. The procedures are as follows. Firstly, K-means was utilized to group pixels in color images into five different color layers and the bounding-boxes of connected-components in every color layer were obtained. Some non-text components were removed by analyzing geometrical properties of the bounding-boxes. Subsequently, the scattered components were connected to form candidate text blocks according to their bounding-boxes＇ geometrical properties and locations. Finally, text verification was performed by geometrical and edge density identifications. The experimental results show that the proposed method is feasible and effective.

关 键 词：场景文本检测 色彩聚类 文本验证 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]