The Research Of Image Mosaic Technology Based On Accurate Registration Of Feature Points

Image Mosaic has a wide range of appplications in computer vision, virtual reality, medical research, military reconnaissance, remote sensing, space exploration. But traditional image mosaic methods have limitations on the efficiency and precision. In this paper, image mosaic based on precise registration with feature points is studied, focusing on the precise image registration based on feature points.In this paper, two kinds of feature points are studied furtherly, one are the corner points, the other are the feature points which have invariant feature. Moravec operator, SUSAN operator, Harris operator and SIFT operator are introduced and a new operator based on Harris operator is presented. To match corner points, SSD, CC, NCC descriptors are introduced, and to match feature points based on SIFT, k-d tree struct and BBF algorithm are introduced. To get rid of the wrong matches, which are called as outliers, RANSAC algorithm is introduced. And In this paper, a new improved algorithm based on RANSAC is presented. At last, a new image fusion method using the Gaussian function is presented to fuse the images.The main contribution of this paper can be concluded as follow. First, a new feature point detection method is introduced, which is accuracy and precise. Second, anlynise the factors which affect the accuracy and precision of the traditional RANSAC algorithm, and this paper proposed an improved algorithm. Third, propose a new method of image fusion which uses the Gaussian function. At last, this paper propose two kinds of image mosaic method based on the above improvements.Through the analysis and comparison of the experiment results, it proves that the new methods of this paper can improve the efficiency and accuracy of image mosaic and they are proved more robustness.