Research Of Remote Sensing Image Fusion Based On Curvelet Transform

Curvelet was developped on the basis of monoscale ridgelet or local ridgelet transform. Compared with wavelet, it has excellent characteristics of multi-scale, spacial domain and frequency domain. It also has multi-direction to represent the edge directions of an image exactly. It best represents the edges and smooth regions of an image in the same reconstruction precision. At present, curvelet transform has been used widely in image processing.This thesis studies some important problems in remote sensing image fusion based on curvelet transform. It includes following aspects.After introducing concept of remote sensing image fusion and home and abroad development situation, it discusses advance procedures for image registration to solve the problem of image fusion being effected by original image.It introduces the theories of wavelet, ridgelet, curvelet and the second generation curvelet transform. Implementation methods and main properties are also mentioned. It does singularity analysis of curvelet transform, discusses how selection of decomposition level to effect remote sensing image fusion. It compares curvelet transform's characteristic and capability by experiment. The result shows that, in the image fusion of an SAR image and an TM multispectral image, curvelet transform keeps nice texture informations of SAR image and spectral characteristics of TM image compared with traditional fusion methods.Because single factor evaluating fusion performance only considered single characteristic of fusion image, global superiority was absent. So this thesis brings forwards new evaluation index to evaluate fusion performance, they are based on spacial detail information and spectral information. The index was uesd in actual evaluation in all kinds of fusion menthods, then there are nicer evaluation standards for algorithms in qualitative and quantitative analysis.It introduces fusion menthod based on culvelet transform, and brings forwards new local LBP operator fusion algorithm based on the second generation curvelet transform. It obtained fine result. This method could keep the spectral information of an original image farthest, at the same time, it enhances the definition and spacial resolution of it. The experiment shows that curvelet transform is better than wavelet transform for remote sensing image fusion.