The Research Of Image Mosaicing Based On Feature Matching Algorithm

With the development of image processing technology,image mosaicing is more and more widely used in the field of aerospace,biomedicine,military security,computer vision,etc.Image mosaicing technique is for getting an image of wide view and high resolution.The main method is to use the camera to shoot the images with overlapping area,and then adopt the computer processing to get a large image.The current research is mainly around the matching accuracy,computation time and removing shadow in transition region.In this paper,not only increasing the speed of calculation but also making the transition region more natural possibly.In the paper,the basic theory and key technology of image mosaicing have been studied,and a method of image mosaicing based on feature matching has been raised.The main contents include following:Firstly,this paper researched a large number of domestic and foreign technology related with image processing,especially studied on image registration and image fusion.Then the related technology of image mosaicing is introduced in this paper,mainly about image collection,image projection model,image transformation model.In the part of image registration,Harris corner detection algorithm,SIFT algorithm and SURF algorithm have been introduced in detail.An improved method of feature matching based on ORB algorithm is raised,which employs the ORB algorithm for extracting binary descriptors of images.Then it adopts median filtering and RANSAC algorithm to get rid of the noise.At last,image correction uses perspective transformation and then it completes image matching.The experiments are compared with SIFT algorithm and SURF algorithm.In the image mosaicing part,two groups of the images with overlapping area are as the experimental object,which obtain an image of wide view and high resolution under the cylindrical projection.In image fusion process,several classical algorithms of image fusion have been introduced,such as PCA(Principal Component Analysis),IHS(Intensity,Hue,Saturation)and PCNN(PCNN-Pulse Coupled Neural Network)etc..In this part,wavelet transformation is the key method.In view of the fusion rule,an improved algorithm based on wavelet transformation is proposed.The algorithm gets high frequency coefficients and low frequencycoefficients by wavelet transformation.While low frequency coefficients cite Sobel operator as its fusion rule,high frequency coefficients cite Brenner evaluation function as its fusion rule.The analysis of results adopts the coefficient weighted average method and the maximum coefficient method as comparative experiments.The objective evaluation methods are the standard deviation(SD),spatial frequency(SF),average gradient(AG)and clarity.