Wavelet-based the Character Recognition in Map

The automatic recognition, including the character, annotation, contour line and so on, is very important in many fields. Here, we focus on the character recognition of the scanned map, because the digitizing map is an important resource data for GIS and automatic recognition character of scanning map is still a key problem, which needs to be solved seriously. In our investigation, which is about automatic recognizing characters in scanning map utilizing zero crossings of wavelet transform, we firstly transformed the gray map scanned to the binary image, which is pre-processed (noise filter, thinning, etc). The closed-contours in the binary image are constructed using morphology operators and the chain codes of the closed-contour are traced. Then, the closed-contours related to the characters are separated from the binary image by the special threshold. The three features (x, y, and tangential angle in the closed-contour of characters ) are extracted, which is of great benefit to change 2D image to 1D signals. The feature signal related to x, y, and tangential angle is decomposed respectively by the discrete dyadic wavelet in 2 scale, and the zero-crossings of the wavelet transform are extracted, because the zero-crossings represent sharp variation points of the feature signals. The dynamic time warping is used as the similarity measure for the purpose of realizing the feature match and character recognition.