Self-adaptive Remote Sensing Image Fusion Based On Empirical Mode Decomposition And Savitzky-Golay Method

Multi-source remote sensing image fusion is a technology of image processing that uses different sensor images or different band images from one sensor to generate a new image. The new integrated image should both have the spectral information and detail information of different sensor images or different band images. With the rapid development of image processing, multi-source remote sensing image fusion has become a hot research and application field in remote sensing. A series of fusion experiments were carried out using Quick Bird and TM multispectral and panchromatic images as the source data based on comparative analysis of the existing traditional fusion algorithms in this paper. The main work and conclusions are as follows:(1) The advantages and disadvantages of traditional fusion techniques such as ISH transform fusion, principal component analysis (PCA) fusion, Brovey fusion, wavelet transform fusion and high-pass filter (HPF) fusion were comparatively analyzed through experiments. The problems that exist in the traditional fusion algorithms were presented.(2) The empirical mode decomposition method was applied to high-low frequency information separation in remote sensing image fusion and a new remote sensing image fusion algorithm based on the two-dimensional empirical mode decomposition was proposed. Firstly, the empirical mode decomposition based on row and column was used in image fusion. As the belt noise was introduced by the new fusion method, a fusion algorithm based on the two-dimensional empirical mode decomposition was used to separate the high-low frequency information. The fusion algorithm based on the two-dimensional empirical mode decomposition can effectively remove the belt noise was proved by experiment. And compared with traditional fusion algorithms, this method can enhance the image detail while maintaining excellent spectral information.(3) The Savitzky-Golay method was extended to two-dimension, and then was used to the high-low frequency information separation. A new remote sensing image fusion algorithm based on Savitzky-Golay method was proposed. A series of Savitzky-Golay operators with different order and power were designed. Then a detailed comparison was carried out between the new fusion algorithm and wavelet transform fusion. The conclusion that the number of the order of Savitzky-Golay operator and the number of decomposition of wavelet transform has the same effect in image fusion was presented. Higher order of Savitzky-Golay operator and bigger of number of wavelet transform would both result in more detail information but less spectral information.(4) The fusion rules that used in traditional fusion algorithms were analyzed and summarized. A new fusion rule for the fusion methods based on two-dimensional empirical mode decomposition and Savitzky-Golay was proposed to make the fusion algorithms self-adaptive. Lastly, the conclusion that the fusion rule could harmonize the detail information and spectral information well was presented.