Abstract: Different approaches to the image feature selection were summarized. In view of the image global features are more likely to be detected and not liable to be disturbed by noise and local distortion, an approach to the image feature selection and extraction based on the global feature was proposed. The mean gray-scale of the image was taken as the major feature value. The image was divided into several blocks, the feature values expressed by 10 numbers were obtained by means of horizontal and vertical projection. To realize the quick real-time recognition of the number, the values of the feature numbers after projection were processed in accordance with the original projective direction. The experimental results show that the numbers in an image with moderate noise can be accurately and quickly recognized by the statistical approach which has a good anti-jamming ability.

Key words: vehicle engineering, vision navigation, automatic guided vehicle, numerical recognition, statistics, feature extraction