Extracting text from images or video streams can be widely used in various application fields. In this paper, a new method to extract text from color images is proposed. First, a cellular decomposition quantization method is adopted to quantize color images. Color and texture features are then integrated to detect homogeneously textured regions. Finally, different classification rules are applied to different types of textured regions individually to identify text. A divide-and-conquer strategy is involved to cope with a text of different fonts and sizes and uniform or nonuniform colored text even when these variations appear in a color image simultaneously. The experimental results are satisfactory.
展开▼
