Improved Mask R-CNN for Content Segmentation of Archival Newspaper Images in Library Collection
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.
Mask R-CNNdigitization of newspapercontent segmentationtarget detection
NETL
NSTL
万方数据
