Study On The Image Registration Algorithm Based On SURF Feature Extraction

Image registration refers to the spatial alignment of two(or more)images of the same target.The process of image registration is called image matching or image correlation.It is the basic and core technology branch in digital image processing and machine vision.It is also the basis of the application of stereo vision,image fusion and dynamic video detection.It has been widely used in the field of medical imaging,remote sensing,military navigation,geographic information system,target recognition,aerospace technology,virtual reality and artificial intelligence.With the complexity of application technology,higher requirements are put forward for the real-time,accuracy,adaptability and high efficiency of image registration algorithm.Gray-based registration algorithm is easily influenced by illumination and gray-scale information,moreover,it has high complexity and low efficiency.When there is nonlinear transformation,the method based on transform domain is limited.While invariant features of images can reflect its essential attributes and can maintain its stability under the condition of certain stretching,rotation,illumination and changing of angle of view.This is of great significance for the study of human image recognition.Therefore,this paper mainly studies the image registration method of SURF feature extraction,and improves the registration accuracy by reducing the interference feature points and purifying the mismatch points.SURF was developed by SIFT.Firstly,the principle,procedure and performance of SIFT and SURF were analyzed and then verified by experiments.The SURF algorithm is more efficient than the SIFT algorithm because it uses box filtering and integral image theory.SURF has higher real-time than SIFT,but there is still the problem of low registration accuracy.Aiming at the existing problems of SURF algorithm,an improved SURF image registration method is proposed.Compared with the traditional SURF,this paper first introduced bilateral filter to the image before feature extraction to reduce the source of error.In the initial phase of feature matching,since the selection of the ratio of nearest neighbor to next nearest neighbor affects the matching result,this dissertation abandons the traditional fixed threshold and uses the adaptive threshold constraint.In order to make the registration more accurate,Kendall Coefficient constraints were added to the matching pairs.Finally,the results were processed by RANSAC algorithm and LSM iteration solution.The experimental results show that the improved SURF algorithm can enhance the correct matching rate based on reducing the registration time.