Research On Algorithm Of Remote Sensing Image Mosaic

In order to expand field of vision and obtain more precise position information in remote sensing image applications, adjacent remote sensing images need to be spliced one image.In this paper, two key techniques of remote sensing image mosaic,which are image registration and image fusion, are focus on.Results of experiments and analysis of several registration algorithms demonstrates that registration algorithm based on SIFT feature has good robustness. So the two adjacent images are sharpened and matched by SIFT algorithm. Then two-way matching algorithm and Random Sample Consensus algorithm(RANSAC) are used to remove false matches.Experiments show that this method can improve the matching rate and also has good registration effect on fuzzy remote sensing image.The image after registration has ghosting and stitching seam in fusion process. So how to eliminate the ghosting and the seam becomes two very important nodes. The common stitching line detection algorithm used to eliminate the ghosting cut the image object mechanically and are lack of ensuring the integrity of the target. So an improved detection algorithm based on dynamic programming is put forward. This method gets a line through gradient values and correlation coefficients in adjacent field of each pixel in overlapping area.Experiments indicate that this method can eliminate the ghosting, bypass the goal edge and avoid cutting object effectively even when the image feature are complex and has large difference in gray-scale. For the common methods to eliminate the stitching seam is poor in applicability when the gray-scale difference is large. A algorithm based on gray difference ratio is proposed. The mosaic image is corrected by the ratio of two values which one is counted by two means of image on both sides of the seam and another is calculated by means of gray value of pixels on both sides of the seam. Results show that this method has a good effect on eliminating stitching seam and retain the texture features of the original image. It has higher information entropy increases respectively by 4.9% and 2.9%, compared withcompulsive correction method and the algorithm based on slope correction ratio. Even when the gray difference is large, the algorithm still has strong applicability.