A Novel Method of Feature Points Identification Using Multi-Scale Technology

High precision identification of feature points is an important technology and one of the bases of computer vision, image analysis and image processing. In the practical applications, the feature points can not be identified easily in various conditions of light illumination, visual angle, texture, and perspective projection. And in many cases, the size of feature point is small, the context is uncertain and the shape of feature point is different from each other. According to the properties of actual vision images, a novel method of feature point identification using multi-scale detecting technology is presented. In this thesis, an automatic circle feature point detection system is developed based on digital image processing techniques, in which, the feature points are detected in two scales. In large-scale recognition, the image pre-processing is carried out to reduce the impact of image noise, and the feature points with various shape deformation can be found out. And then, the founded points are checked along with small scale detecting. The relationship of the points and their mappings in three dimensional space is obtained and used to judge if the point is a real feature point one by one. Thus the deformed feature points in various perspective projection conditions can be recognized correctly. Experimental results show that this approach for feature points identification is not only fast but also robust. The proposed technology is a kind of useful method for contour extraction and feature points identification in the fields of computer vision and image analysis.