基于MSER视频字幕敏感词过滤算法  

Video Subtitle Sensitive Word Filtering Algorithm Based on MSER

作　　者：李雅静 丁海洋 LI Yajing;DING Haiyang(School of Information Engineering,Beijing Institute of Graphic Communication,Beijing 102600,China)

机构地区：[1]北京印刷学院信息工程学院,北京102600

出　　处：《现代信息科技》2023年第21期80-84,89,共6页Modern Information Technology

基　　金：国家自然科学基金(61370188);北京市教委科研计划(KM202010015009,KM202110015004);北京印刷学院科研创新团队项目(Eb202101);北京印刷学院重点教改项目(22150121033/009);北京印刷学院科研基础研究一般项目(Ec202201);北京印刷学院博士启动金项目(27170120003/020,27170122006)。

摘　　要：当前网络视频中充斥着涉及敏感信息的内容且不易识别,提出了一种基于最大稳定极值区域和字典树的视频字幕敏感词过滤算法,达到对此类视频内容过滤、处理的目的。该方法采用MSER算法对文字进行检测与定位,来获得候选文字区域,然后通过使用几何和笔画宽度属性来减少这些可能区域的数量,再经过几何过滤的方法对候选区域进行筛选、优化,连接候选区域以获得文本区域,最后利用字典树算法进行敏感词过滤。实验结果表明,该方法可以有效地对视频字幕敏感信息进行检测和过滤。At present,network videos are full of content involving sensitive information and are not easy to identify,and a video subtitle sensitive word filtering algorithm based on the maximum stable extremal region and dictionary tree is proposed to achieve the purpose of filtering and processing such video content.This method uses MSER algorithm to detect and locate text to obtain candidate text regions,and then reduces the number of these possible regions by using geometry and stroke width attributes.After the screening and optimization of candidate regions by geometric filtering method,it connects candidate regions to obtain text regions,and finally uses dictionary tree algorithm for sensitive word filtering.Experimental results show that this method can effectively detect and filter video subtitle sensitive information.

关 键 词：最大稳定极值区域 文字检测 视频图像 文字定位 敏感词过滤 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]