Panchromatic And Multispectral Image Fusion Algorithms Based On Local Features

With the rapid development of remote sensing technology, the remote sensing images can be obtained with different sensors loaded in the satellite. Multispectral images, obtained by multispectral sensor, have high spectral resolution but low spatial resolution, while panchromatic image, obtained by panchromatic sensor, is a gray image with high spatial resolution. According to both multispectral images and panchromatic image, a multispectral remote sensing images with high spatial resolution can be obtained by the image fusion technology. In this paper, the fusion algorithms of the panchromatic image and multispectral one are researched for the spatial resolution with2:1and4:1ratios. The research goal of this thesis is to improve the spatial resolution and reduce the spectrum loss.For the spatial resolution2:1of the panchromatic image and multispectral images, the existing fusion algorithms are studied and simulated. The analysis of this thesis shows that the multispectral fusion image got by existing fusion algorithms has a issue of spectral distortion or little detail information. To overcome these problems, a pan-sharpening algorithm for remote sensing image is proposed based on local correlation. Firstly, the amplified multispectral images and the original panchromatic image will be decomposed by NSCT with the same number of layer. And then the local correlation coefficients and fourth-order correlation coefficients can be computed in the low frequency sub-band of the decomposed images. If the value of the local correlation coefficients is greater than that of fourth-order correlation coefficients, the corresponding high frequency coefficients of multispectral images will be replaced by the panchromatic image’s, otherwise, they should be kept in same. Lastly, the inverse NSCT is used to reconstruct the fused image according to the low frequency coefficients of multispectral image and high frequency coefficients obtained by the fusion procedure. Experimental results show that the algorithm not only improves the spatial resolution, but also preserve the spectral characteristics of multispectral images.While the spatial resolution of the panchromatic image and multispectral images equal to4:1, the4times magnification multispectral image obtained by the general interpolation methods will cause the spectrum loss of the fusion results. Therefore, this paper proposes a fusion algorithm based on dictionary training and regional characteristics. The design of this algorithm is focalized on the amplification method and the fusion rule of high frequency. First of all, the pan-sharpening algorithm based on dictionary training is used to amplify the low-resolution multispectral images. This makes the amplified multispectral images keep more spectral characteristics of original images, and reduce the spectrum loss of fusion results. Then, the low frequency coefficients of fusion image equal to the corresponding coefficients of multispectral images based on dictionary training to retain more spectral information, while the high frequency coefficients of it are obtained by the fusion based on regional variance matching to enhance the detail and texture information. At last, the inverse NSCT is used to reconstruct the fused image according to the low frequency coefficients of multispectral images based on dictionary training and high frequency coefficients obtained from the fusion procedure. Experimental results show that the algorithm enhance spatial resolution of multispectral images and also retain better spectrum for the panchromatic and multispectral images with a space resolution ratio4:].