A new method to capture the feature vectors of numeric alphabetic based on circles, lr- sides contour and row subsection is proposed by intergating the opinions of Statistic Feature Extraction and Structural Feature Extraction. It uses less feature vectors to keep the substantial information in the topologic structure of a numeric alphabetic, which has a very strong adaptability. In the imitating experiment, first the feature vectors extracted are to be assorted roughly based on the Structural Features (circle) of the numeric alphabetic, and then the other feature vectors is to be recognized and eliminated gradually making use of the AD AdaBoost neural network. The result shows that, this method not only has a faster speed of recognition, but also has a higher ratio of correct recognition.