Research On Remote Sensing Image Fusion Method

Remote sensing image fusion is to generate a composite image from multi-images of different sensors or different scale images of same sensor using a certain algorithm.It can improve the clarity and legibility of image objects and get the characteristic information that original image is not available.The fusion image is widely used in terrain rendering,Agricultural surveys and military defense and other fields.The experimental data is band B2,B3,B4 and B8 of US Landsat satellite images which cover Lijiang River of Guilin.The spatial resolution of band B2,B3 and B4 is very low with 30m,but the spectral resolution is high with combined color image.However,the spatial resolution of band B8 is very high with 15m,but the spectral resolution is low with grey image.The fusion image from band B2,B3,B4 and B8 with certain fusion algorithm will be high spatial resolution of 15 m and high spectral resolution of color image.The main methods of this paper are weighted sum Method,IHS fusion Method,PCA fusion Method and wavelet transform fusion Method.The fusion algorithms were achieved in computer language programming.The fusion images were evaluated with subjective vision and objective indicators such as average gradient,edge intensity,entropy,mean gray,standard deviation,correlation coefficient,distortion.The fusion images were got with weighted sum Method,IHS fusion Method,PCA fusion Method.The optimal fusion weights of weighted sum Method is 0.5 and 0.5.High resolution B8 full color image and low resolution multi-spectral B2,B3,B4 image IHS transform brightness component,PCA transform the first principal component of the histogram matching,matching the results instead of low resolution multi-spectral image IHS transform brightness Component,PCA transforms the first principal component.Then the fusion images were evaluated with subjective vision and objective indicators.The results show the best fusion effect is with IHS fusion Method in which the ground objects are more recognizable.The fusion effect of PCA fusion Method is slightly lower,but the color of image fusion is very abundant.The fusion image of weighted sum Method is relatively poor.High frequency coefficients and low frequency coefficients were set different weights in wavelet fusion Method and the fusion images were evaluated with subjective vision and objective indicators.The results show the best fusion effect is with average of low frequency coefficients and maximum of high frequency coefficients.The ground objects in fusion image with maximum of low frequency coefficients and maximum of high frequency coefficients is very clear,but the values of objective indicators is not stable.The fusion image with average of low frequency coefficients and average of high frequency coefficients is the worst.The fusion image with regional variance matching degree in wavelet fusion Method has a good spatial resolution and spectral resolution when the threshold is 0.75.The ground objects such as karst hills,urban streets and building are very clear and distinguishable.The fusion images with wavelet regional variance matching degree Method is better than those with wavelet coefficient weighted Method,IHS fusion Method,PCA fusion Method.The results can be used to identify ground targets more accurately in remote sensing image,provided services for the application of remote sensing images.