Research On Image Mosaic Based On Feature Points

The real world is a rich data source, among which, visible data is the most important. An image retrieved by image mosaic is an effective way to represent the real world. The images from image mosaic cleaned up redundant information among images sequences, compressed information sizes, thus better represented the information. More important, the real world is shown more objective and vivid. With the development of computer science, wide angle and high resolution images or videos are playing important roles in some fields. Image mosaic solved the problem caused by wide angle and resolution, and thus provides a good approach for those fields.Therefore, the image mosaic is widely applied to photogrammetry, computer vision, remote sensing image processing, medical image analysis, computer graphic and so on. The thesis first introduces the present research status of image mosaics technology, and the application domain of mosaics, thus has demonstrated the broad application prospect of image mosaics technology. Then the thesis discusses the basic steps of image mosaic, image acquisition methods and models of image mosaic. The common image registration methods and image fusion methods are introduced, and their performance as well as the existing problems is analyzed. And the method of image mosaic based on feature points is introduced especially. Then this thesis proposed a new image mosaic algorithm(CSM,Curvature Scale-space Mosaic). The Curvature Scale Space is employed to detect the feature points. And the matching point pairs are obtained by comparing the Euclidean distance of the Zernike moments, which are computed around a circular neighborhood centered at the feature point. Then the transform parameters are computed through the correct mapping model. Finally, the weighed average method is used to achieve image mosaic. Experimental results demonstrate that the proposed algorithm can effectively stitch two images which are different by rotation, translation and noises.