Research On Remote Sensing Image Mosaic Technology Based On Features

Digital image stitching technology refers to generating a panoramic image with great view from a group of images which have overlapping regions, according to the corresponding relationship between the overlapping regions. Remote sensing image is generally used in large area of regional studies, and because of the high altitude location restrictions, it usually requires several all-round shots, and then mosaics two or more than two remote sensing images.The remote sensing image mosaic has wide use demand, providing timely and effective information basis for geographical environment monitoring, military surveillance, the control of natural disasters and so on.The thesis mainly focuses on applying the feature-based image stitching technology to the remote sensing image, putting forward some improvements on feature points extraction and image registration, then discussing the further optimizing of sequences image registration, finally achieving the seamless fusion.The main work of this paper is as follows.1. For high resolution digital remote sensing image, an improved feture point extractions algorithm based on SURF has been presented in this paper. Firstly, this paper uses edge detection to retain strong features as image preprocessing. Then, it accelerates the process of calculating high resolution images’transforms relationship by estimating the overlapping regions in advance. Experiments show that the algorithm can effectively suppress the number of feature points, and make the points’distribution more rational. Furthermore, the detection and matching process only be made on the overlapping region effectively improve execution speed, and make the subsequent mosaic more smoothly.2. According to the fact that digital remote sensing image has many texture similar regions, this paper proposes an improved matching algorithm based on traditional SURF feature nearest neighbor matching algorithm. It use gray domain and spatial domain self-correlation to restrain the preliminary registration result. Experiments show that the improved algorithm can retain most correct matching point pair, and eliminate false ones, thus effectively improve the accuracy of image matching.3. In the process of sequence images stitching, this paper researches the correct matching between images and cumulative error problem, using an iterative method to optimize the transform matrix between two images, making the global error minimization. Considering the ghost effect of remote sensing image fusion result, use graph cut based best seam search technique and weighted average way to eliminate the ghosting phenomenon.4. In this paper, we design and implement a digital remote sensing image processing system. In the system’s stitching module, we realize the algorithms mentioned above, and provide different process approaches based on two images and multiple images.