改进Mask R-CNN的馆藏报纸图像内容分割  

Improved Mask R-CNN for Content Segmentation of Archival Newspaper Images in Library Collection

作　　者：倪劼[1] 叶江松 谢恩泽[1] Ni Jie;YE Jiangsong;Xie Enze

机构地区：[1]南京图书馆

出　　处：《图书馆论坛》2024年第6期110-118,共9页Library Tribune

基　　金：江苏省图书馆学会研究课题“基于深度学习的民国时期报纸数字化关键内容研究”(项目编号:22YB067)研究成果。

摘　　要：开展馆藏报纸图像内容分割研究,能提升文字识别准确率,对促进机器识别取代人工操作、提高图书馆数字化工作效率具有重要意义。文章根据报纸图像呈现的特征,提出一种基于改进MaskR-CNN的算法,实现报纸图像内容分割。首先,通过优化锚框比例和损失函数,对原始MaskR-CNN算法进行改进。其次,采用数据增强、调整训练参数开展样本训练。最后,通过实验的方式对改进后的MaskR-CNN算法训练模型和原始算法训练模型进行比较,并采用AP_bbox和AP_segm评价指标对实验结果进行评估,改进后的算法训练模型AP_bbox为0.935,AP_segm为0.943,均超过原始算法训练模型。实验结果表明,改进后的MaskR-CNN算法能够实现报纸图像内容有效检测与分割。The study of content segmentation of newspaper images in the library collections is crucial for enhancing the accuracy of text recognition,which is of great importance for advancing machine recognition to replace manual operation,and improving the efficiency of digitization in libraries.This paper proposes an algorithm based on an improved Mask R-CNN to separate the content of newspaper images.First,the original Mask R-CNN is improved by optimizing the anchor box ratios and loss functions.Secondly,samples are trained by data augmentation and adjusted training parameters.Finally,the training model based on the improved Mask R-CNN is compared with the original model through experiments,and the experimental results are evaluated using the AP_bbox and AP_segm evaluation indicators.The improved algorithm training model scored 0.935 in AP_bbox and 0.943 in AP_segm,outperforming the original training model in both categories.This study suggests that the improved Mask R-CNN algorithm can achieve effective detection and segmentation of newspaper image content.

关 键 词：Mask R-CNN 报纸数字化 内容分割 目标检测 

分 类 号：G210.7[文化科学—新闻学] TP183[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]