Research On Corner Detector And Image Matching Based On Corner

One important step in the technique of image processing is matching. It can be divided into tow class: the area method and the feature method. The normalized cross correlation algorithm is a in common use method of the area matching. Since in the case of dissimilar image matching, the matching probability of correlation algorithm is not good. Recent many studies are focus on feature matching.Feature matching include tow step, one is the select of feature and feature detect, another is the matching measure. Improved approach is described in this paper, it, based on the 'corner' feature and 'Hausdorff distance', adopt 'genetic algorithm' in real-time matching. Analyze point feature and compare other popular detectors, we introduce a new fast and efficient corner detector based on 'SUSAN' and the thought of corner structure. In order to reduce time, First pre-detect is applied, then based the result of pre-detect the corner structure is analyzed, that is meaning that if a pixel is a corner, it must accord with corner structure. It is commonly considered that corners are image points that show a strong two dimensional intensity change, and are therefore well distinguished from near points. So that salient function is used as corner response function to get high information.Improved average Hausdorff distance is advanced in this paper, and it is used in to image matching. The application that improved Hausdorff distance used in image feature matching is implemented. Experiments have shown that when the tow images' gray are dissimilar, feature matching based on corner is better than normalized cross correlation algorithm. In some fields, corner matching is valuable and utility, to optimize matching process, genetic algorithm is introduced into real-time calculation. Many experiments are implemented both on FCD corner detector and feature matching suggested in this paper. Location accurate and use time and corner number of FCD is calculated and compare to other popular detectors. Experi ments have shown that the new corner detector is accurate and efficient, can be use to real-timeapplications. Matching between image obtained by 'Indian planet' and spot image is done, result shows that our method is better than the normalized cross correlation algorithm when the images are dissimilar and contain enough information.