An Edge-based Approach for Sign Text Extraction
Liang Dong · Yingxiang jishu · 2006
A sign has various kinds of important information and therefore,sign text extraction in natural scene images is very useful in many applications.In this paper,we present an effective edge-based method of text extraction.The proposed method is robust to background complexities,color,size and alignment of texts.Firstly,many irregular long edges are eliminated.Secondly,a multi-level method is performed according to different ranges of character sizes.In each level,we scan the edge image using a window to further remove non-text edges.Thirdly,results at different levels are merged together.High accuracy is achieved by considering the characteristic of text edges during the stage of verification.We illustrate experimental results on a large number of images.