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 ON VORONOI DIAGRAM AND CONDITIONAL RANDOM FIELD
方炳坤 1楚瀛2
作者信息
- 1. 上海电力大学电气工程学院 上海 200090;上海电力人工智能工程技术研究中心 上海 200090
- 2. 上海电力大学电气工程学院 上海 200090
- 折叠
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％.
Abstract
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Key words
Bayesian model/Conditional random field/Voronoi diagram/Computer vision/Text detection引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据