Research On Image Matching Algorithm Based On Scale Invariant Local Feature

Image matching is one of the important research contents in the field of computer vision,and it is also the research foundation of many popular issues.It is indispensable in the fields of target location,visual navigation,3D reconstruction,image mosaic and image recognition.The purpose of image matching is to find the correspondence between the images obtained at different times,different angles and different shooting conditions.In order to obtain better image matching effect,how to eliminate or reduce the error factors in the matching process and extract the characteristics of high stability has become the focus of research in related fields.Therefore,this dissertation improves the SIFT algorithm by using Nonsubsampled Contourlet transform(NSCT),fuzzy enhancement algorithm and edge feature and color information of image.The main work of this dissertation is as follows:(1)In order to eliminate or reduce the influence of the error factors in the matching process on the SIFT feature matching performance,a matching preprocessing method based on NSCT and fuzzy enhancement is implemented.Firstly,NSCT decomposition is carried out on the reference image and the image to be matched respectively by using the flexibility of NSCT decomposition,and the low frequency components are decomposed to suppress the influence of high frequency noise.And then performs image enhancement processing on the low frequency component to enhance the image texture information.Finally,the SIFT feature is extracted on the image component after the above steps to perform the pre-matching,and the mismatched pair is removed by the random sampling consistent RANSAC algorithm.(2)The SIFT feature transformation method based on edge feature and color information is implemented for the local feature extracted by the SIFT algorithm can not divide the local gray scale similarity region of the image.Firstly,the Canny operator is used to obtain the edge set of the image,and the color feature of the corresponding pixel is calculated.Then,the key points are described by using the idea of shape context.The main steps are to establish the logarithmic polar coordinate system and divide the feature point circle neighborhood into 60 regions with SIFT feature points,and then counts the color accumulation of the edge points in each region.Finally,the extracted feature and the KL transformed local gradient feature are combined to form a linear cascade descriptor,and a reasonable measure criterion is matched for the feature descriptor.(3)The effectiveness of the pretreatment method and the feature transformation method implemented in this dissertation are verified by experiments,the algorithm improves the matching performance of the original algorithm under the complex transformation of rotation and scale change,illumination transformation,angle of view transformation and image blur.