Research Of Character Recognition Algorithm In Meteorological Facsimile Map Vectorization

Meteorological facsimile maps as a kind of weather map can provide meteorological information in a more simple and intuitive way. However, the dot-matrix expression nature of the meteorological facsimile map limits the breadth and depth of its application. Therefore, many scholars devoted to the research of meteorological facsimile map vectorization. The vectorized meteorological facsimile map can both eliminate the influence of artificial intervention and promote the superposition use with other meteorological information.Meteorological facsimile maps include much meteorological information, such as the latitude and longitude lines, coastlines, isobars and annotations. The algorithm of recognizing annotations in vectorizing meteorological facsimile map is the key research in this thesis. Annotations mainly include the symbols indicating the air pressure center and the numerical figures around the isobars. The first step of the vectorization is the preprocessing of the dotted matrix map, including the elimination of base map, the binarization for the image and the detection of longitude lines through Hough transformation.The next step is the extraction of the symbols indicating the air pressure center. This step includes the preliminary extraction of the symbols indicating the air pressure center, the template matching based on correlation between the identified character and the template character and to remove the influence of isobars. The final step is the research for the algorithm of recognizing annotations around the isobars. This step mainly includes the character detection, the character segmentation and the character recognition. According to the degree of characters’deformation as well as the interference of surrounding isobars and other information, the algorithm uses the hierarchical processing thought to study the character recognition. The character detection and segmentation in each level is different according to the projection features. The character recognition in each level uses the shape context algorithm. In addition, the character rotation plan is studied, the influence for the character during the artificial process is made up and the priori knowledge is used to correct the recognition results. Experimental results show that employing the algorithms presented in this thesis, annotation symbols on the meteorological facsimile map can be extracted and recognized successfully.