Performance Improvement Of Meteorological Facsimile Map Vectorization Method

Meteorological facsimile maps are pictures which can reflect weather conditions in a certain area. Through meteorological facsimile maps, ships sailing on the sea can know the weather evolution process about more and more navigation areas, master the weather and the maritime situation which has occurred and will occur on navigation areas, to ensure the safety of maritime activities. But meteorological facsimile maps, which are showed in bitmap form, not only occupies large storage space, but also not conducive to processing and superimposed display, which limits its breadth and depth of application. Therefore, a lot of scholars are dedicated to the study of the vector of the meteorological facsimile maps.Meteorological facsimile maps include much meteorological information, such as the latitude and longitude lines, coastlines, isobars, frontlines and annotations. Improving the algorithm of annotations recognition to improve the recognition rate and speed, and the study on algorithm of extracting, recognizing and vectorizing frontline is the key research in this thesis. The first step of the vectorization is to improve the algorithm of recognizing annotations around the isobars, mainly including preliminary extraction characters, character detection and segmentation by character projection, character recognition by center point shape context. Due to the character distortion or the effect of isobar and other lines on annotations, when character detection and segmentation, possible segmentation points need to be predicted according to segmentation points which have been found, testing whether possible segmentation point are segmentation points. The final step is the study for the algorithm of extracting and recognizing frontlines, mainly including preliminary extraction of frontline, the study on line tracking algorithm based on a-O filter principle, removing interference lines based on contour tracing, frontline identification recognition based on template matching, correcting recognition errors by relationship of the front line and identifications. Because frontline recognitions are on frontline, frontline will not be thinned, when tracking front line, the front line center is determined by seeking centroid, and broken line is connected by the line trend and breakpoints distance.Experimental results show that algorithms presented in this thesis can improve the performance of character recognition, and frontlines can be better recognized and vectorized.