Document zone classification using machine learning

When processing document images, an important step is classifying the zones they contain into meaningful categories such as text, halftone pictures, line drawings, and mathematical formulae. A character recognition system, for example, may confine its attention to zones that are classified as text, while in an image compressor may employ specialized techniques and models for zones such as halftone pictures. The penalty for incorrect classification may range from incorrect interpretation to reduced efficiency. In any case, the higher the classification accuracy, the better the results.