Current Result Document :
| 한글제목(Korean Title) |
다중 스케일 기능 향상 네트워크에 기반한 임의 형태의 장면 텍스트 감지 |
| 영문제목(English Title) |
Arbitrary-shaped Scene Text Detection based on Multi-scale Feature Enhancement Network |
| 저자(Author) |
My-Tham Dinh
이귀상
My-Tham Dinh
Guee-Sang Lee
|
| 원문수록처(Citation) |
VOL 49 NO. 01 PP. 0669 ~ 0671 (2022. 06) |
한글내용
(Korean Abstract) |
|
영문내용
(English Abstract) |
Scene text detection is an exciting and vital topic of scene text reading systems. It has been widely applied in recent years for natural scene image understanding, visually impaired people reading the text, etc. The arbitrary-shaped is the main challenge of scene texts. In this paper, we propose a novel framework - a Multi-scale Feature Enhancement Network (FEN) with each FEN able to help model enhance the receptive fields and representation capabilities of the extracted features from a lightweight backbone. Accordingly, a Multi-scale FEN is learned deeper by three-scale convolutional kernels 3x3, 5x5, and 7x7 so that the arbitrary-shaped text in natural images has been tackled more effectively. This experiment achieves a competitive F-measure of 82% on ICDAR 2015 dataset. |
| 키워드(Keyword) |
Scene Text Detection
Deep Learning
Feature Enhancement Network
Arbitrary-shaped
Segmentation.
|
| 파일첨부 |
PDF 다운로드
|
