Text detection in images using texture feature from strokes

Abstract

Text embedded in images or videos is indispensable to understand multimedia information. In this paper we propose a new text detection method using the texture feature derived from text strokes. The method consists of four steps: wavelet multiresolution decomposition, thresholding and pixel labeling, text detection using texture features from strokes, and refinement of mask image. Experiment results show that our method is effective. © Springer-Verlag Berlin Heidelberg 2006.