Fusion Based Seamless Mosaic for Remote Sensing Images

Abstract

A fusion based seamless mosaic method for stitching remote sensing images is introduced in this paper. The proposed method focuses on one major problem in the process of mosaic: how to generate visually pleasant stitching result in the cases of misalignment, global and local intensity differences between images. First, two partially overlapped images are decomposed into high-frequency components and low-frequency components with Gaussian low-pass filter. Second, by considering the information characteristics contained in both separated components, different mosaic schemes are designed to accomplish stitching process accordingly. For the low-frequency components consisting of coarse shape and illumination information, two-dimension weighted blending rule is utilized to achieve smoothing transition. For the high-frequency components including rich details, an improved seam searching strategy based on dynamic programming is introduced. With the obtained stitching seam guiding the stitching process, visible structural break can be avoided. Finally, the mosaic result is produced by linearly composing both stitching results of different components together. Experimental results demonstrate the effectiveness of the proposed method in generating seamless mosaic results without introducing any unexpected blurring or artifacts.