基于视觉显著性和提升框架的场景文字背景抑制方法  

Visual Saliency and Boosting Based Background Suppression for Scene Text

作　　者：姜维[1] 卢朝阳[1] 李静[1] 刘晓佩[1] 姚超[1] 

机构地区：[1]西安电子科技大学综合业务网国家重点实验室,西安710071

出　　处：《电子与信息学报》2014年第3期617-623,共7页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(60872141);中央高校基本科研业务费专项资金(K50510010007);华为高校创新研究计划(IRP-2012-03-06)资助课题

摘　　要：为解决复杂背景对场景文字自动定位算法干扰的问题,该文利用视觉显著性抑制背景且突出前景的特点,以方向梯度直方图特征、方向梯度直方图统计特征、梯度幅度特征和梯度曲线特征的弱分类器,结合提升框架提出一种背景抑制算法。该文算法的目标是抑制自然图像中复杂背景且突出前景文字,作为场景文字自动定位算法的预处理阶段增强算法效果。在ICDAR2011场景文字定位竞赛数据库和实验室场景中文数据库中实验结果表明,该文算法较好地抑制自然场景中复杂背景,并有效提升场景文字自动定位算法的性能。To solve the issue of background interferences on the scene text automatic localization algorithm, a scheme of background suppression for scene text is proposed, which utilizes characteristic of visual saliency to combine histogram of oriented gradient features, its statistical features, gradient magnitude features and gradient curve features with the boosting frame. The scheme aims to suppressing the complex background and highlighting the foreground text in natural scene. It can consider be as the preprocessing stage of the scene text automatic localization algorithm, and it improves the performances of the scene text automatic localization algorithm. The experimental results in both the ICDAR2011 scene text localization competition test dataset and the laboratory Chinese dataset show that the proposed scheme can suppress effectively the complex background and improve the scene text localization algorithm.

关 键 词：图像处理 场景文字 背景抑制 视觉显著性 提升方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]