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

With the rapid development of new sensors, the ability of people to obtain images also extremely improved. While, images received from a single kind of image data or single sensor often can't satisfy the actual needs. Image mosaic refers to synthesis of two or more images which have the overlapping area and to describe the same scene by generating a new image which has larger angle range of this scene. Image registration and image fusion are the two key techniques of image mosaic, they have a very close relation to each other. Image registration is the base condition for image fusion. In order to realize the precise image fusion, we must do the geometric registration of images first. Image registration and image fusion technologies are widely used in medicine, remote sensing, aerospace, virtual reality and other fields, which are so important and can't be replaced by other technologies.The thesis summarized the relevant technologies of image mosaic based on features. Image mosaic based on features includes the following steps:extracting features, matching features, estimating model parameters, transforming image. Based on methods in each step of image mosaic, the thesis emphasized on the relevant algorithms of the image registration based on SIFT features and the image fusion based on wavelet transform.In order to realize image registration, the thesis first introduced image transformation model and then emphasized on SIFT (Scale Invariant Feature transform) which is an algorithm of feature detection and has no invariance on rotating, dimension zoom and brightness. In the match to SIFT feature, we adopted the following two strategies to get the accurate matching points:the first one is to get initial matching points from these feature points by the matching method of similarity criterion in Euclidean distance; the other one is to get accurate matching points by the improved RANSAC to eliminate mismatch points.On the basis of accurate registration, the thesis outlined the theories and methods of wavelet transform, and applied it to image fusion. Fusion rules, which determine the quality of the merged image, are the core of image fusion based on wavelet transform, so the thesis focused on studying of low-frequency and high-frequency fusion rules, and proposed an improved low-frequency fusion rule which was based on windows and related to the correlation coefficient and the spatial frequencies. The thesis also described the methods to evaluate the quality of fused images.The thesis used the subjective and objective evaluation indicators of image fusion quality to evaluate the fused images. The experiment results showed that the fused images have higher quality when using the method proposed in this thesis.