Research On Feature Points Algorithm For Imagine Matching

Image matching algorithm is an important and meaningful research of image processing. The feature matching of image have many advantages, such as small amount of calculation, good ability of the changes of gray scale, noise, affine transformation and block, so it has been widely used in practice.Scale-invariant feature matching (SIFT) is an image feature point matching algorithm based on the amount of image invariant featues. It have a very good role in scale space, image scaling, rotation and affine transformation, but in the actual application process, there is still room for improvement. In this paper, NMI and IMED are used to improve the SIFT. As the NMI algorithm match, it has the advantages of good resistance to deformation geometry, high robustness, simple calculation and so on, it is used in SIFT algorithm to simplify the process of computing SIFT and improve the efficiency of the alogrithm. The IMED algorithm fully considered the relevance of spatial pixels, more objcetively reflects the same and differences between the matrix. It is used in the SIFT, fully considered the relevance of SIFT descriptor, effectively improve the accuracy of the alogrithm.A number of simulation experiments shows that, under the different conditions of rotation, noise, perspective, blur, compression, light, the SIFT-IMED have the best accuracy, the SIFT and SIFT-NMI have the similar accuracy, and the caculated efficiency in descending order is SIFT-NMI, SIFT, SIFT-IMED. Experimental results consistent with the theoretical results, and achieve the desired goals.