2019년 컴퓨터종합학술대회
Current Result Document :
| 한글제목(Korean Title) |
멀티 스케일 퓨전 네트워크를 이용한 문서영상 이진화 |
| 영문제목(English Title) |
Document Image Binarization with Multi-scale Fusion Network |
| 저자(Author) |
Quang-Vinh Dang
이귀상
Guee-Sang Lee
|
| 원문수록처(Citation) |
VOL 46 NO. 01 PP. 0820 ~ 0822 (2019. 06) |
한글내용
(Korean Abstract) |
|
영문내용
(English Abstract) |
Binarization of degraded document images is an important pre-processing step for document image analysis domain. We develop a ladder net with a multi scale structure to learn text-like features from document images itself to classify text and background from degraded document images. Specifically, we consider two properly designed LadderNet architectures: one with deeper architectures, another with shallower architecture. Each structure is trained independently using document image patches. The target of our design is to predict the foreground maps at two different feature levels. Predicted maps have fewer noises in the background from the shallower architecture with larger size of striding window. However, the detail of the text is not clear. On the other hand, predicted maps generated from the deeper architecture with smaller size of striding window have clearer text strokes but contain more background noises. Therefore, a better result is achieved by combining the outputs of two these architectures. The proposed approach achieves state-of-the-art results on DIBCO datasets, revealing the robustness of the presented method. |
| 키워드(Keyword) |
|
| 파일첨부 |
PDF 다운로드
|
