Automatic panoramic stitching of omni-directional remote sensing images

Panoramic stitching of multiple high-resolution remote sensing images still faces the accuracy and the speed difficulties. This paper proposed an automatic panoramic stitching algorithm to register multiple omni-directional remote sensing images, which utilizes Location Sensitive Hashing (LSH) to accelerate the rough feature matching, and combines the Euclidean distance ratio and the median filter to achieve the stable homographic transform. The decision of the reference image from the multiple images is judged by the most sum of the overlapping matches density. Experimental results show that the proposed method can get faster registration speed than the iterative method under the comparable accuracy.