earticle

영어

Due to the overlapped region and image size of the input image pair is unpredictable, it makes the matching procedure more difficult and unstable. For the purpose of finding out the stable and accurate matching algorithm of large images, we give an analysis of different kinds of characters, such as, scale invariant feature transform (SIFT), local maximum gradient descriptor, Harris corners, and the maximum curvature points of the image edges, etc. Based on the experiments of different images, a new stable matching algorithm is proposed in this paper. In our model, the matching procedure is divided into two stages, the rough and accurate matching procedures. To evaluate the matching result, the edge information is combined with the local maximum gradient of the input images as the constraint of our matching algorithm. After the extraction of the local maximum gradient character points, we use the edge information to divide these points into different classes. Then, the searching of the stable and accurate matching problem becomes to find out the best matching results which agree to the edge constraints. The experimental results show that the proposed algorithm is more efficient and stable than the other kinds of matching algorithms especially in the proposing of large size images.