Research Methods Of Image Registration Based On SIFT And Image Fusion Based On WAVELET

Image mosaic can realize to two or more images splicing processing, which contain the scene was the overlapping area,and then makes the scene view in a large radius.Although many image matching algorithms which have different implementation ways,their key technologies are to realize image registration and image fusion.Image registration is to establish the corresponding relationship between the two images,and image fusion is to mosaic transition more smooth and prevent possible joint which produced by light differences.Image mosaic technology is applied widely,For example,they are used in artificial intelligence robot navigation,in large 3D virtual reality game,in battlefield reconnaissance warning and panoramic view of military action and in the Internet + pathological image of remote medical consultation, etc.The thesis introduced the image mosaic technology systematically,and analyzed the basic process of image mosaic.For two key technologies: image registration and image fusion, the thesis did the further detailed analysis.During a presentation on image registration techniques,through the following three aspects.Firstly,the thesis introduces the related basic theoretical knowledge, such as common transformation model classification;secondly,Second, the thesis introduced the property of registration, which affected by three basic framework.Thirdly,the thesis introduced two kinds of classification of image registration,and introduced the process of image matching based on feature, which was adopted by the paper.Next in image fusion technology, the two points were introduced.On the one hand, the thesis introduced the related concepts of image fusion,which consisted of three different levels.On the other hand, the thesis introduced the two methods based on pixel level fusion, including the article adopted in the frequency domain by wavelet transform fusion method.In the process of image registration, in order to achieve a better match results, the paper adopted based on SIFT registration algorithm,and the improved method was proposed from that.Therefore, the paper revolved around the image registration steps have been introduced in detail,for the SIFT feature extraction, the feature matching and the image transformation relationship to solve.Then, the following two strategies were put forward,One was the eigen-vector matching process had used a linear combination of the chessboard distance and blocks and got the optimal value for image matching instead of Euclidean distance, to improve the computing speed and reduce the matchingtime; After the initial match, in order to eliminate the false matching and improve the matching accuracy, the thesis adopted the RANSAC(random sampling uniformity)algorithm, and then used to calculate the matching point after purification of projection transformation matrix parameters.After the implement strict registration,based on wavelet transform was used to realize the image fusion in the thesis.The paper introduced the wavelet technology detailedly,and introduced a fast algorithm of wavelet decomposition and reconstruction,which been came up with by Mallat.the thesis also focused on the fusion rules, and then put forward an improved fusion rules.Finally, a brief introduction to the quality of image fusion evaluation criterion.