基于灰度波动的非均匀光照文档图像二值化  

BINARIZATION OF UNEVEN ILLUMINATION DOCUMENT IMAGE BASED ON GRAY LEVEL FLUCTUATION

作　　者：王刘奎 史伟民[1] 杨亮亮 Wang Liukui;Shi Weimin;Yang Liangliang(Faculty of Mechanical Engineering and Automation,Zhejiang Sci-Tech University,Hangzhou 310018,Zhejiang,China;Mstar Technologies,Inc.,Hangzhou 311121,Zhejiang,China)

机构地区：[1]浙江理工大学机械与自动控制学院,浙江杭州310018 [2]杭州汇萃智能科技有限公司,浙江杭州311121

出　　处：《计算机应用与软件》2020年第12期197-203,共7页Computer Applications and Software

基　　金：国家重点研发计划项目(2017YFB1304000);浙江省自然科学基金项目(LY18E050016,LY18E070006)。

摘　　要：光学字符识别的算法研究中,快速有效地将文档图像二值化是图像预处理阶段的关键步骤。在实际拍摄中,由于环境复杂,往往使得采集的图片出现光照不均匀的现象,严重影响后续处理。针对该问题,提出一种基于灰度波动的邻域局部阈值分割方法。根据目标考察点一定邻域范围内水平和竖直方向的灰度波动曲线,提取所有的波峰点和波谷点信息,经过运算确定当前阈值。实验结果表明,该方法处理效果远远优于Niblack、Sauvola等算法,与经典的Wellner法几乎没有差别。该方法不仅细节保留较好,无断笔伪影现象,且适应不同光照情况。In the research of optical character recognition algorithm,fast and effective binarization of document image is the key step in image preprocessing stage.However,in actual shooting,because of the complex environment,the collected images often show uneven illumination,which seriously affects the subsequent processing.To solve this problem,a local threshold segmentation method based on gray fluctuation is proposed.According to the gray fluctuation curve of horizontal and vertical direction in a certain neighborhood of the target inspection point,all peak and trough points were extracted to determine the current threshold.The experimental results show that the processing effect of this method is much better than that of Niabck and Bernsen algorithms,and it is almost no different from the classical Wellner method.This method not only retains the details well,but also has no artifact of continuous pen,and can adapt to different illumination conditions.

关 键 词：非均匀光照 文档图像 二值化 灰度波动 局部阈值 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]