Study On Visibe Spectral Remote Sensing Image Mosaic Based On Feature Points

Due to the restriction of the image-acquisition conditions, the images of the large senceshave to be got by mosaicing of related images. Among the techniques of image mosaic, thegeometrical registration technique can influence the accuracy and visual effect of mosaicingresults. Now, as one of the main image registration methods, the image registration methodbased on feature point is not only applied to the registration between images with lesseroverlapping, but also used on the images contain active scene and some defiladed objects. Thispaper took the image registration algorithm based on feature point as the main line, andin-depth researched the image mosaicing technology that grounded on SIFT feature pointmatching. Thus, it explored an effective kind of mosaic technologies for remote sensing opticalimages .The mainly researching work in this paper includes the following:1. It summarized and analyzed development status and the problems of the image mosaictechnologies in detail, identified the main researched contents and summarized theimage-connecting and the image transformation models.2. It systematically studied the basic principles and related several matching methods ofHarris feature point, including, the square of pixels difference, cross-correlation information ofpixels, normalized cross-correlation information, etc. It analyzed and compared theperformance and disadvantages of each image matching method. Based on the systematicalanalysis of the SIFT feature point matching algorithm, an improved SIFT feature imageregistration algorithm is presented. With the invariant property of rotation and scaletransformation in SIFT features, the problem of sensitive to the scale change in Harris featuresdetection algorithm is effectively improved in this paper, which maked image registration underlarge scale transformation to be possible.3. To the problem that the accumulated errors during the mosaicing process may influencethe quality of the achieved image, an overall optimization adjustment method based on featurepoint is presented. It firstly select and sequence corresponding points, thereby selectively todetermine the corresponding points that used to explanted the factors of the connecting models.With the guarantee of the accuracy of the explanation, it can reduce the explanting timeeffectively.4. In this paper, image mosaic software is developed. It can complete independent tests forcharacteristic point extraction arithmetic operator and geometric mosaicing of multiple imageswith features overlapping.