To solve the problem of adaptability and accuracy in the field of image feature extraction and feature matching, the method of scale invariant feature transform (SIFT) was introduced. Firstly the characteristics of the SIFT method were analyzed by theory, and the SIFT method was implemented step by step on the visual studio 2005 platform; Then the features extracted by SIFT method were applied to match images on the criterion of nearest neighbor based on Euclidean distance. A suggestion value bound was given by comparing the matching result of different nearest ratio. At last the effect of the SIFT method was validated by the matching result of three different groups of images. The matching result shows that the features extracted by SIFT method are invariant to image scale, rotation and illumination change, and the matching accuracies are all equal or close to 100%. These results prove that the features extracted by SIFT method have excellent adaptive and accurate characteristics, which are useful for the fields of image recognition, image reconstruction, etc.
Lowe D G. Distinctive image features from scale-invariant keypoints[J].International Journal of Computer Vision.2004, 60 (2):91-110
Witkin A P, Babaud J, Baudin M. Uniqueness of the Gaussian kernel for scale-space filtering[J].IEEE Transactions on Pattern Analysis and Machine Intelligence.1986,8(1):26-33
Witkin A P. Scale space filtering Proceedings International Joint Conference Artificial Intelligence. Karlsruhe,Germany:CA, 1983:1019-1022
Koenderink J J. The structure of images[J].Biological Cybernetics.1984,50:363-396
Lindeberg T. Detecting salient blob-like image structures and their scales with a scale-space primal sketch: a method for focus-of-attention[J].International Journal of Computer Vision.1993,11(3):283-318
Lindeberg�� T. Scale-space theory: a basic tool for analysing structures at different scales[J].Journal of Applied Statistics, 1994,21(2):224-270 ��
Mikolajczyk�� K. Detection of local features invariant to affine transformations .France:Institute National Polytechnique de Grenoble, 2002