Research Of Automatic Remote Sensing Image Mosaic Technique

This paper discusses about Automatic Remote Sensing Image Mosaic technique,which creates high resolution images with a very wide view or full view panoramicmosaics from two images or image sequences automatically. These images may betaken in different time, from different perspectives or by different cameras.This paper presents a novel approach to mosaic aerial images automatically fromimage sequences taken in flight strips and between these strips, which is based onfeature points. There is also an improved solution of image mosaic based on SIFTpresented in this paper.In this new image mosaic approach, we first extract Moravec feature points, thendo coarse matching by Normalized Cross Correlation (NCC) method based on thesepoints. In this part several searching strategies were tested and compared. Then LeastSquare Registration (LSR) method is introduced to refine these matching points. Wecan also gain transform parameters of the geometry transformation model in thisprocess. After image registration, all these images are converted to one unified imagecoordinate system, and the overlapped parts between images are processed byweighted average method. At last, we use the nearest domain method to deal with thepixel coordinates sampling interpolation problem. For every step in this approach,there is another kind of solution for comparison given in this paper, and both theadvantages and disadvantages of every two solutions are verified by experiments. Itturns out our approach can reach a better result with a higher efficiency.We also use LSR to replace RANSAC as a registration improvement in Imagemosaic algorithm based on SIFT, for the latter may fail sometimes. Because of its highconvergence speed and the property that it takes every data into account, LSR methodnot only guarantees the accuracy of registration, also makes the correct rate of thealgorithm improved, and also reduces the running time of the algorithm.