融合背景估计与U-Net的文档图像二值化算法  

Document image binarization algorithm based on background estimation and U-Net

作　　者：熊炜[1,2] 王鑫睿 王娟[1,2] 刘敏[1,2] 曾春艳[1,2] Xiong Wei;Wang Xinrui;Wang Juan;Liu Min;Zeng Chunyan(School of Electrical&Electronic Engineering,Hubei University of Technology,Wuhan 430068,China;Hubei Collaborative Innovation Center for High-efficiency Utilization of Solar Energy,Hubei University of Technology,Wuhan 430068,China)

机构地区：[1]湖北工业大学电气与电子工程学院,武汉430068 [2]湖北工业大学太阳能高效利用湖北省协同创新中心,武汉430068

出　　处：《计算机应用研究》2020年第3期896-900,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61571182,61601177);湖北省自然科学基金资助项目(2019CFB530);国家留学基金资助项目(201808420418)。

摘　　要：针对低质量文档图像存在页面污渍、墨迹浸润、背景纹理等多种退化因素,提出一种融合背景估计与U型卷积神经网络(U-Net)的文档图像二值化算法。该算法首先进行图像对比度增强,然后通过形态学闭操作来估计文档图像背景,并利用全卷积网络,即U-Net对背景减除图像进行前景背景分割,最后采用全局最优阈值处理方法获得最终二值图像。实验结果表明,在2016和2017年国际文档图像二值化竞赛(DIBCO)中该算法的F值(F-measure,FM)、伪F值(pseudo F-measure,p-FM)、峰值信噪比(peak signal to noise ratio,PSNR)、距离倒数失真度量(distance reciprocal distortion,DRD)比性能次优的经典算法最高有5.58%、2.47%、0.86 dB、1.19%的性能提升。Low quality document images have various degradation factors,such as page stains,ink bleed-through and background texture.This paper proposed a novel document image binarization algorithm based on background estimation and U-Net.The algorithm first performed image contrast enhancement,and estimated the document image background via morphological closing operations.It then adopted a fully convolutional network,namely the U-Net,to extract the foreground text from the document background.Finally,it used the global optimal thresholding method to obtain the resulting binary image.Experimental results show that in the 2016 and 2017 DIBCO the proposed method outperforms other state-of-the-art document image binarization algorithms in terms of F-measure,pseudo F-measure,PSNR and DRD,with 5.58%、2.47%、0.86 dB and 1.19%.

关 键 词：文档图像二值化 对比度增强 形态学闭操作 U型卷积神经网络 全局最优阈值处理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]