基于Voronoi图与条件随机场的自然场景文本检测方法  

TEXT DETECTION IN NATURE SCENE IMAGE BASED ONVORONOI DIAGRAM AND CONDITIONAL RANDOM FIELD

作　　者：方炳坤 楚瀛 Fang Bingkun;Chu Ying(Shanghai University of Electric Power,Shanghai 200090,China;Shanghai Electric Power Artificial Intelligence Engineering Technology Research Center,Shanghai 200090,China)

机构地区：[1]上海电力大学电气工程学院,上海200090 [2]上海电力人工智能工程技术研究中心,上海200090

出　　处：《计算机应用与软件》2024年第1期119-125,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(51807114)。

摘　　要：在自然场景中准确有效地检测文本是一项艰巨的任务,故提出一种基于条件随机场(CRF)框架的场景文本检测方法。通过利用贝叶斯推断估计文本极大值区域的置信度作为一元成本项,通过使用维诺图(Voronoi图)来构建CRF空间邻域信息,从而构建图模型,通过最大流算法最小化成本函数区分文本与非文本标记;利用字符的几何特性通过聚类方法聚合成行。实验结果表明,该算法比传统基于最大稳定极值区域(MSER)算法性能有所提高,自然场景文本检测正确率能达到87%。It is an arduous task to accurately and effectively detect text in natural scenes.A scene text detection method based on the conditional random field(CRF)framework is proposed.By using Bayesian inference to estimate the confidence of the text maximum region as a unary cost item,by using the Voronoi diagram to construct the CRF spatial neighborhood information,the graph model was constructed.The maximum flow algorithm was used to minimize the cost function to distinguish text from non-text Mark.The geometric characteristics of the characters were used to cluster them into rows.The experimental results show that the proposed algorithm is improved compared with the traditional MSER algorithm,and the accuracy rate of natural scene text detection can reach 87%.

关 键 词：贝叶斯模型 条件随机场 VORONOI图 计算机视觉 文本检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]