Image Fusion Algorithm Based On Weighted Sum And Structural Similarity

Since the Soviet Union launched its first artificial satellite in 1957 and harvested the first image of the Earth in 1959,Earth observation has entered people's field of vision and opened the door to remote sensing research.With the continuous advancement of technology,multi-temporal and multi-resolution images are constantly emerging,and images produced by a single sensor are increasingly unable to meet the ever-increasing demands of a wide range of applications.Therefore,fusion methods of different sensor images have emerged.The fusion of different sensor images,combined with the advantages of different types of images,can effectively improve the usage rate of a single image,and tap more useful information to achieve complementary information between the original images.The merged image is more conducive to the subsequent research work of remote sensing image information extraction and feature recognition.In this paper,we explore the fusion algorithm of GF-2 images,select a single highresolution image of the ground object as the fusion test image,test the fusion effect;finally extend the fusion method to the feature type.Enrich or highlight the GF-2 images of different features.The fusion process is divided into two parts: firstly,multi-spectral and panchromatic images are processed using histogram equalization and recursive filtering methods for initial image fusion.Then the multi-scale geometric transformation is combined with the structural similarity adjustment weighting strategy to complete the weighted fusion,and the final result is obtained.for evaluating the method of this paper,the method of this paper is compared with the results of common fusion methods.Finally,the applicability of this method to different types of remote sensing images is discussed.The research in this paper includes the following:(1)Analyzed the existing problems in remote sensing image fusion,and studied the three levels of remote sensing image fusion.According to the fusion process,the advantages of different fusion levels were compared and analyzed.Disadvantages.It focuses on the analysis of several common pixel-level fusion methods,such as the Brovey method,the principal component transformation method,and the theoretical basis of the wavelet transform fusion method.(2)The first fusion of images.The appropriate pre-processing of the image to be fused is performed to obtain multi-spectral and pan-color band images of different sensor types of the same size.In the initial fusion process,most of the images are directly weighted and averaged,and the threshold is artificially selected,the original image histogram is first equalized and then median filtered.Then,the equalized image and the median filtered image are compared to obtain a local color difference,and the color difference is combined with the brightness to be normalized and marked.The marker image and the local contrast are multiplied to perform weight estimation,and the recursive filtering refines the weight estimation and then inverse transform to obtain the first fusion image.(3)Image secondary fusion.In the initial fusion results,the resolution does not fully reflect the advantages of the original panchromatic image.For this deficiency,the first fusion image and the two original images are first subjected to Non-subsample Contourlet Transform(NSCT).It is decomposed into high frequency information and low frequency information.The local brightness,local contrast and local contrast parameters are calculated for the low frequency information,and the structural similarity is calculated according to these three parameters.Then,the similar area classification identification map is classified to calculate the weight,and the low frequency information is fused by the weight,and the high frequency image is fused according to the correlation intensity.Finally,the final fused image is obtained by inverse transform of NSCT.(4)Quality evaluation.The fusion results of this paper are compared with the fusion results of the Brovey method,principal component transformation method and wavelet transform method to obtain subjective visual evaluation.By calculating six evaluation indicators,including average gradient,image sharpness,information entropy and structural similarity,cross entropy and relative standard deviation,quantitative evaluation.(5)Exploring the method proposed in this paper,whether it has wide applicability to different feature types of high-resolution images.It also fused visible and infrared images,and tests the fusion effects,and evaluates them both subjectively and objectively.