Paper Info
Title
Multi-lingual Scene Text Detection Based on Fully Convolutional Networks.
Abstract
In the paper, we propose a method based on transfer learning to detect multi-lingual text in natural scenes. First, a semantic segmentation map of the input image is obtained through a fully convolution network (FCN). In this map, each pixel is classified to text or none-text. And then, the candidate boxes of text regions are computed based on the map. In this procedure, VGG network is trained to obtain a basic character classifier of single language. Based on this VGG model, FCN has the ability to classify each pixel to text or none-text for multi-lingual with doing transfer learning. Finally, the bounding boxes of text are carry out by filtering the unsatisfied candidates with some rules. The experimental results show that our method achieves good performance on the task of multi-lingual text detection. And compared with other advanced method, the time cost of our method is shortest.
Year
DOI
Venue
2017
10.1007/978-3-319-77380-3_40
ADVANCES IN MULTIMEDIA INFORMATION PROCESSING - PCM 2017, PT I
Keywords
Field
DocType
Sence text detection,Multi-language,Transfer learning,Fully convolution networks
Computer vision,Pattern recognition,Segmentation,Convolution,Computer science,Transfer of learning,Filter (signal processing),Pixel,Artificial intelligence,Classifier (linguistics),Text detection,Bounding overwatch
Conference
Volume
ISSN
Citations 
10735
0302-9743
0
PageRank 
References 
Authors
0.34
16
6
Name
Order
Citations
PageRank
Shaohua Liu121.39
Yan Shang200.34
Jizhong Han335554.72
Xi Wang401.69
Hongchao Gao502.70
Dongqin Liu600.34