Research On A New Method For Image Registration Based On SIFT

Image registration is one of the key techniques in image application and processing.It has been widely used in many procedures of image processing such as image mosaic,environmental detection,image fusion,precision guidance localization and images change detection.With the research of image application becoming more thorough,more and more requirements are demanded in the efficiency,stability and adaptability if image registration technique.Scale Invariant Feature Transform(SIFT)feature point detection algorithm can extract scale and rotation invariant image features which also remain their invariance under varying illumination and vision,thus being applied in remote image registration.Whereas,since the SIFT algorithm is based on local neighborhood gradient information,it will be invalid when there exists a great contrast difference between the pictures and the gradient information directly obtained from the pixel difference values is instable.In response to the above-mentioned problems,this thesis proposes a new image registration method based on edge strength graph and scale invariant feature.This method can be described as the following three steps:first,phase congruency information will be utilized to establish the edge strength graph in order to eliminate the impact of image contrast difference;second,scale invariant feature points detection will be done in the edge strength graph with the descriptors of these feature points obtained;finally,new matching strategies will be put forward as solutions to the mismatching problem of SIFT,and image registration will be completed based on the precise matching points obtained.In this study,an experiment will be carried out on many images with great contrast difference chosen from SAR images from different remote sensors,multispectral images in varied wave bands,SAR images and optical images.The stimulation results show that the registration method in this thesis is an effective and practical solution to the registration problem of images with great contrast difference,operation time being 4～5 seconds,with high registration precision and excellent rotation resistance.