Image Registration Based On Feature Extraction

Image registration is an important research direction in the field of digital image processing,and it is also a difficult point in digital image processing.It is widely used in most aspects of social life.Image registration is the operation process of two or more images obtained under different times or scenes to achieve the aim of image alignment through a mapping relationship.Finally,image registration obtains the geometric transformation parameters that can reflect the mapping relationship between images.Image registration algorithm based on feature extraction usually has two important steps: image feature extraction and feature matching.The features of image can be roughly divided into three categories: point feature,line feature and surface feature.In this thesis,we extract the edge feature(line feature)of image as the feature of image registration.The edge of image is caused by the drastic change of the gray value in the image,and it is a widely existing in the image.After analyzing the edge features of the image.In this thesis,we modify the multi-pixel anisotropic Gaussian filter which is an effective edge detection operator.A multi-scale-based MAGF edge detection operator is proposed.The new algorithm adds scale space on the basis of the original MAGF,and can extract the edge information at various scales,thus avoiding the loss of edge details and the impact of noise interference on the detection results.Owing to the speckle noise in the SAR image is different from the additive noise model in the ordinary optical image,the edge detection operator,which has good performance in the ordinary optical image,is not suitable for extracting the edge features of the SAR image.Therefore,after analyzing the characteristics of SAR images,this thesis proposes a modified ratio-based MAGF edge detection operator based upon MAGF.First,the filter mask is divided into positive and negative two parts,and then filter the image after take the absolute value of two new masks.Finally,calculating the ratio which used as the edge detection variable of SAR image.The edge features of SAR image can be extracted more effectively.Experiments show that the improved edge detection operator can extract the edges of SAR images more effectively.In terms of feature matching,the shape context as a registration algorithm to describe the shape features of objects,has attracted great attention since it was put forward.However,there are only one kind of evaluation criterion adopted by the shape descriptors,the number of feature points that fall into the bins.After the analysis of the possible distribution of the feature points in a bin,it is found that one single evaluation criterion does not distinguish feature points very well.Therefore,in the improved shape context algorithm,the evaluation criterion of shape descriptor is increased to three,the number of feature points that fall into the bin,the center and the variance of those feature points.Through simulation experiments,it is proved that the improved shape descriptor can distinguish most of the shape feature points,and the registration rate and the average error are better than the shape context algorithm.