基于局部对比度和相位保持降噪的古籍图像二值化算法  

ANCIENT DOCUMENT IMAGE BINARIZATION BASED ON LOCAL IMAGE CONTRAST AND PHASE PRESERVING DENOISING

作　　者：冯炎 Feng Yan(School of Information Science and Technology,Tibet University,Lhasa 850000,Tibet,China)

机构地区：[1]西藏大学信息科学技术学院,西藏拉萨850000

出　　处：《计算机应用与软件》2022年第2期162-166,173,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(61661047)。

摘　　要：图像二值化算法通过消除文档背景噪声将文本与背景分割开。针对古籍图像提出一种基于局部对比度和相位保持降噪的古籍图像二值化算法。根据归一化局部最大值最小值来构造古籍图像局部对比度,同时对古籍图像进行相位保持降噪。将局部对比度图像和降噪图像相结合来识别文本笔划像素。通过局部窗口内所检测的文本笔划像素估计局部阈值从而计算古籍背景修复模板。用图像修复算法和形态学闭操作来估计古籍背景。用所估计背景来增强古籍图像,采用Howe算法对增强后的古籍图像进行二值化求得最终结果。该算法在DIBCO2016、DIBCO2017和DIBCO2018数据集中的实验结果均优于其他二值化算法。Document image binarization can segment the document image into background and text by removing the degradations.This paper proposes an ancient document image binarization based on local image contrast and phase preserving denoising.A local image contrast map was constructed by the normalized local maximum and minimum,and phase preserving denoising was performed.The local contrast map was combined with denoised image to identify the text stroke pixels.The background inpainting mask was further calculated by a local threshold that is estimated based on the detected text stroke within a local window.The background was estimated with image inpainting procedure and morphological closing operation.The ancient document image was enhanced with the background,and the Howe algorithm was used to binarize the enhanzed ancient document image to obtain the final result.The experimental results on the DIBCO2016,DIBCO2017 and DIBCO2018 data sets show that the proposed method achieves top performance compared with other.

关 键 词：局部对比度 相位保持降噪 古籍图像 二值化 图像修复 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]