Research On Remote Sensing Image Fusion Algorithn Based On Directional Multi-resolution Analysis

At present, most earth remote sensing satellites can provide both panchromatic image with high spatial resolution and multi-spectral images with low spatial resolution. Image fusion techniques can effectively combine the high spatial resolution information of panchromatic image with the spectral information of multi-spectral images into one color image. The fused image can be better applied to ground objects identification, land use and land cover survey, and other remote sensing applications.This thesis was concerned primarily with remote sensing image fusion algorithms based on directional multi-resolution analysis. Some directional multi-resolution analysis tools such as discrete wavelet transform, wavelet packet transform, contourlet transform, wavelet-based contourlet transform (WBCT) and non-sampling contourlet transform (NSCT) were used to decompose the original images, and the fusion coefficients were chosen according to different fusion rules. The fusion results were evaluated by some objective evaluation indexes.The main work of the thesis is as follows:(1) We fused the SPOT panchromatic image and multi-spectral images, ALOS panchromatic image and multi-spectral images, SPOT panchromatic image and TM multi-spectral images based on traditional wavelet transform fusion algorithm and analyzed the different fusion results when using different wavelet functions or decomposing the images to different levels. The results showed that, the decomposition level in 2 or 3 is appropriate, and different wavelet functions didn't have significant impact on fusion results. We drew the same conclusions when fusing images based on IHS and wavelet transform combined algorithm.(2) When using the same wavelet function and decomposing the original images to the same level, the spectral deviation of the fusion image which based on IHS and wavelet transform combined algorithm is less than the one based on traditional wavelet transform, but the entropy is approximate.(3) An IHS and wavelet packet transform combined fusion algorithm was proposed. The experiment results showed that, this algorithm can reduce the spectral deviation and improve the correlation coefficient and structural similarity when compared with IHS and wavelet transform combined fusion algorithm. But the spatial information was less than the later. (4) An IHS and wavelet transform integrated fusion algorithm based on SSIM (Structural Similarity) was proposed first. In this algorithm, the low frequency coefficients of new intensity component were defined as the weighted average of the panchromatic and the intensity component, and the weight value was determined adaptively by juding the SSIM between them. The experiment results showed that this algorithm can preserve the spectral characteristics of the multispectral image better, while improving the resolution of the image.(5) In IHS and contourlet transform combined fusion algorithm, different pyramid filters and directional filters were selected when decomposing the original image. We drew the conclusion that filter combination of "5-3" pyramid filter and "pkva8" directional filter can obtain the optimal fusion result.(6) Via Comparing the fusion results those based on directional multi-resolution analysis (wavelet transform, contourlet transform, WBCT, NSCT), it showed that the fusion results based on contourlet and its expansion forms have higher spatial resolution, where the boundaries of linear features such as roads were demonstrated more clearly and the block phenomena was avoided availably.(7) A region-based IHS and WBCT combined fusion algorithm was proposed. At present, most fusion rules were only based on single pixel, and with little regard for the neighborhood characteristics. In our algorithm, the high-frequency coefficients of fusion image were determined by comparing the standard deviations of source images, and the low frequency coefficients were determined by the energy matching degree of source images. It showed that this algorithm can improve the spatial resolution of fusion image, while effectively maintaining the spectral characteristics.