An Improved Method Of Image Registration Algorithm Based On SIFT

Image registration is the basis of the image recognition, In terms of pattern recognition, computer vision, medical image processing and remote sensing information processing have a wide range of applications. Prior to closing date,domestic and foreign scholars have been considerable research for image registration methods, Apparent from research, Image registration can be divided into manual and automatic registration into two categories, Wherein the automatic registration can be divided into based on region and characteristics two areas. Based on region use location information and pixel area, similarity calculation to be registered images and reference image,to get the transformation model parameters to achieve registration.Otherwise, based on feature need to match the image feature extraction, parameters of the model transformation according to the results of matches,that can make image registration achieve robust. After a lot of research experiments, Feature-based image registration technique is superior based on region image registration techniques,so it has been widespread concern of scholars. In this paper, we use the feature-based image registration technique.In the field of image recognition also produced a variety of image registration algorithm, In many algorithms, scale invariant feature transform(SIFT) feature has rotational invariance, scale invariance, significantly, a large number of changes in perspective as well as for stability and noise, These features make the SIFT features after being presented has become a classic image local feature,now, The method in object recognition, 3D reconstruction, video retrieval and other fields has been widely used.SIFT(Scale Invariant Feature Transform)algorithm is based on local invariant feature extraction algorithm, Designed to eliminate image punctual mismatching points.This paper analysis the SIFT algorithm,analysis mismatching sources, Erroneously match point sources to develop programs based on culling.And compared with the traditional registration results of SIFT.In this paper,beginning with SIFT algorithm matching.Then,uses spatial domain feature points excluding mismatching points, Finally, do the final debugging using RANSAC algorithm.Experimental results show that the improved SIFT algorithm can better eliminate mismatch points. D.G Lowe introduced in 1999, and made a perfect and summarized in 2004, Image registration is a key step in image processing,With the gradual deepening of the application of adaptive image matching,matching accuracy and timeliness of the increasingly high demand. This paperbased image registration principle?...