Detecting Text in Image with Distance Transform and Directional Information
Xiaolei WANG
School of Management Huaibei Normal University, Huaibei, Anhui, 235000, CHINA
Abstract: A new method to detect directly text strokes in image are proposed in this paper. Firstly the edges in image are extracted by Minimum-Spanning-Tree-like(MST-like)algorithm, the distance and directional information from the distance transform combined with low gradient amplitude threshold are applied to determine a candidate strokes region. Finally extraction of real strokes are finished by the hierarchical clustering algorithms and some other relevant features of text. Experiments showed that the method of this paper can be improved by the recall rate and precision rate of text detection, for handwritten Chinese text in the image this method also has a good effect.
Keywords: Text Detection; Distance Transform; Directional Information; MST-like; Hierarchical Clustering