Document Image Segmentation into Text, Continuous-Tone and Screened-Halftone Region by the Neural Networks.