摘要
针对在背景复杂、视角多变的仪表图像中提取数字区域信息的需求,本文提出一种改进的仪表数字区域定位方法。本文提出的方法主要是对仪表图像的连通域进行检测、分析和筛选,最终确认数字区域。具体方法为首先对光照不均、模糊的图像进行图像灰度化、直方图均衡、图像滤波、图像二值化等方法进行图像预处理。然后采用MSER算法的准确检测数字区域,产生一系列嵌套的最大极值稳定区域(MSER),对这些区域采用区域清理和区域像素拓展进行优化,建立数字候选区域,然后以笔画宽度变换(SWT)作为文本提取的关键特征,实现对文本候选区域的准确筛选,从而最终确定数字区域。
In order to extract the data from the digital area in a sophisticated, visual image, this paper proposes an improved instrument digital region localization method. The proposed approach is primarily to test theinstrument connected domain of the image, analysis and screening, finally confirm the digital area. in tins paper, the image preprocessing is carried out for image image gray-scale, histogram equalization, image filtering and image binarization. Then, the MSER algorithm is used to accurately detect the digital region and produce a series of maximal maximal stability zones （MSER）, which can be optimized by regional cleaning and regional pixel expansion, and establish a digital candidate region, and then use the stroke width transformation （SWT） as the key feature of text extraction, to realize accurate screening of the text candidate region, and finally determine the digital region.
作者
徐珊
吴开超
张晓丽
Xu Shan;Wu Kaichao;Zhang Xiaoli(Computer Network Information Center, Chinese Academy of Sciences, Beijing 100190, China;University of Chinese Academy of Sciences, Beijing 100190, China)
出处
《科研信息化技术与应用》
2017年第5期19-25,共7页
E-science Technology & Application
基金
自然科学基金项目(71671178)
关键词
数字图像识别
数字区域定位
最大极值区域
笔画宽度变换
digital image recognition
digital location
maximum extremum region
stroke width transform
