深層学習による文書画像の領域分割およびラベル生成ツールASLA における領域分割の精度向上と適用範囲の拡大のための拡張

DOI 機関リポジトリ 機関リポジトリ (HANDLE) オープンアクセス

書誌事項

タイトル別名
  • Extension to Improve the Performance of Region Segmentation and Expand the Scope in ASLA Which Is a Segmentation and Labeling Tool for Document Images Based on Deep Learning

この論文をさがす

説明

The current situation of the electronic documents is only a substitute for paper. As a new way to utilize electronic documents, we focus on dividing electronic documents into regions by their elements and generating keywords and sentences as labels from the contents of the elements. However, these tasks, when performed manually, are timeconsuming and labor-intensive. To reduce the time required for region segmentation and label generation, we developed a prototype of ASLA, a tool for region segmentation and label generation of document images using deep learning in our laboratory. However, the existing ASLA has some problems in terms of accuracy of region segmentation and is not highly useful. Therefore, this study expands the existing ASLA to improve its usefulness. Specifically, first, a rule-based region redividing process is added. And then, the object detection method used in segmentation of ASLA is modified. Furthermore, the dataset is added. As application and evaluation results, we have confirmed that the usefulness of the expanded ASLA is improved.

収録刊行物

キーワード

詳細情報 詳細情報について

問題の指摘

ページトップへ