Text Location for Scene Image with Inherent Features.

Qiaoyu Sun,Yue Lu
DOI: https://doi.org/10.1007/978-3-642-33506-8_64
IF: 8
2012-01-01
Pattern Recognition
Abstract:Locating the text from nature scene images is a challenge problem for computer, because text is various. In this paper, the inherent characteristics of the text are employed to locate the text from complicated background in the scene images. The edge map of the input image is obtained by the Canny operator which is masked by the Sobel edge map. Then the connected components analysis is used to detect the potential characters. Thirdly, the characters in the same line will be merged into a candidate text region. Finally, the symmetry of edge direction and the stroke width of each candidate text region are calculated to verify if it is a true text region. The experimental results on various scene images have demonstrated that the proposed method is capable of effectively locating text regions of the nature scene image.
What problem does this paper attempt to address?