Feature Matching. Three-dimensional Reconstruction Study

Different image matching of the same scene is a key problem in computer vision, and is frequently used in 3-D model reconstruction, object recognition, image alignment, camera self-calibration and so on. This paper is main about the feature extracting and matching techniques of figure picture.In the technique of picture feature extraction, we import B-Spline model with shape parameter, and propose a method to extract prominent point using it. Considering the curve shape adjusted by the parameter. First, the figure fitting by B-Spline model with shape parameter is presented, then the prominent points could be determined from the curvature distribution. In this paper, we conclude the policy of how to choose the shape parameter. Choose the equitable parameter for different figure, could improve the precision of fitting and reduce the spending. The method in this paper could be used widely, and more flexible.In the researching of feather matching, we propose a new method to matching the prominent points. Based on that the prominent points are order, We import the total curvature ,match the prominent points by match the prominent segments. In the process of matching, analysis and improve the total curvature, make sure that it adapt to all kind edges. Experimental results show the validity of this method.