Using 2D tensor voting in text detection
Toan Dinh Nguyen, Jonghyun Park, Guee-Sang Lee · 2010
A novel text detection algorithm based on 2D tensor voting is proposed. Tensor voting is used to extract text line information by exploiting the curve saliency value and curve normal vector at each character. The text line information is useful information to improve the results and reduce the effect of using heuristic rules of region-based methods. The experimental results attained from several natural scene images show that the proposed method successfully detects text with low false positive rate.