Research On Remote Sensing Image Fusion Based On Nonsubsampled Contourlet Transform

Remote sensing image fusion recombines the complementary information of multiple sensor images to produce a composite image with more accurate and richer information,and multispectral-panchromatic image fusion are the most widely studied fusion in remote sensing image fusion processing.Due to the difference of sensor imaging principle and the environment of remote sensing satellite,people often can't get the remote sensing image with high spatial resolution and multi-spectral,which limits the scientific research potential and application prospect of remote sensing data.The fusion of multispectral image and panchromatic image can improve the fusioned images' spatial-resolution while retaining the spectral information of multispectral image,and make full use of the spatial and spectral information of remote sensing image,improve the comprehensibility and clarity of remote sensing images,strengthen the visual interpretation of the RS data,utilize the completed advantages of various remote sensing sensors,and make up for the shortcomings of the remote sensing system.Based on the analysis of various pixel-level remote sensing image fusion methods,this paper focuses on multi-scale decomposition methods--nonsubsampled contourlet transform(NSCT)with the characteristics of multi-scale,multi-direction and translation invariance,and different fusion rules are established for the difference between the low-frequency coefficients and the high-frequency coefficients,and two remote sensing image fusion algorithms based on NSCT are proposed:S_NSCT algorithm--the spatial frequency and regional energy are introduced to formulate the fusion rule of sub-band coefficients.The spatial frequency represents the overall activity of the image,and the regional energy can effectively highlight the detailed texture of the image and ensure the sharpness of the image.So we propose improved NSCT fusion algorithm which low-frequency coefficients is determined by the maximum of spatial frequency and high-frequency coefficient determined by Regional energy weighting;P_NSCT algorithm--PCNN model has the characteristics of global coupling and pulse synchronization,combines the spatial frequency and regional energy to participate in the formulation of sub-band coefficient fusion rules,It can makes full use of the correlation of image feature information,and the fusion result has small spectral distortion and clear spatial detail structure.so,an NSCT fusion algorithm based on spatial frequency and regional energy impulse PCNN model is proposed.In order to verify the practicability of the two remote sensing image fusion algorithms proposed in this paper,four different remote sensing data are selected as the source data of the fusion experiment,including: landsat8 satellite images,GF-1 satellite images,"JL-1" optical-A images,SPOT and TM multi-source rs images,and NSCT-PCNN algorithm,NSCT algorithm,IHS/PCA algorithm and Gram-Schmidt algorithm are selected for comparison analysis.The experimental results show that the average gradient(AG)value of each band in the image fusion by S_NSCT algorithm is the largest,and the information entropy(EN),correlation coefficient(CL),standard deviation(SD)of each band in P_NSCT algorithm are the largest,while the spectral distortion WR is the smallest.The results show that the proposed remote sensing image fusion algorithm has more image information,less spectral distortion and highly image sharpness,which effectively improves the quality of multispectral-panchromatic fused images.The fusioned image based on S_NSCT algorithm has more spatial details,while The fusioned image based on P_NSCT algorithm has more spectral informatio.