Research On Matching And Fusion Methods For Multi-source Remote Sensing Images

In the field of remote sensing, satellite remote sensing systems are providing a growing number of global coverage and repeated measure data. To make full use of this information, multi-source images of the same target or the same region need to be matched and fused. In this paper, multi-source remote sensing images are taken as the research objects, the matching and fusion methods for multi-source remote sensing images are studied and implemented.Firstly, a fast image matching algorithm based on contourlet-domain, Krawtchouk moments and improved particle swarm optimization is discussed, which is mainly used to match the remote sensing images taken from a single sensor. Compared with those of other existing image matching methods, the algorithm is less time-consuming, while greatly improves the precision.Secondly, aiming at the remote sensing images taken from multi-sensors, two image matching algorithms are studied. Improved hausdorff distance and Tsallis entropy based mutual information are used as matching measure function, respectively. The results show that the algorithms both have the higher accuracy and require fewer operations.Then, an image registration method based on log-polar transformation and alignment metric is implemented. Log-polar transform is applied to the region whose centre is Harris corner, and the matching points are obtained by using alignment metric. The results show that the method can effectively match the multi-source remote sensing images, and has robustness with rotation and scale transform.Next, the nonsubsampled contourlet transform is used in image fusion for multi-source remote sensing images. The approach can achieve better results than the methods based on wavelet transform or contourlet transform.Finally, an image fusion method based on nonsubsampled contourlet transform and fuzzy reasoning is introduced, estimating the importance of the source images with fuzzy reasoning. The results show that the method is simple, feasible and can be more effective for image fusion.