Research On Technology Of Multi-source Remote Sensing Image Fusion

The rapid development of remote sensing technology provides a kind of effective technical means for the human being to know the living environment and utilize the natural resources.There are a variety of remote sensing sensors,and the images formed by different image sensors have distinct characteristics,resulting in such as multi-spectral images,panchromatic images,infrared images,synthetic aperture radar(SAR)images and other multi-source remote sensing images.In order to efficiently and comprehensively deal with these image data,the multi-source remote sensing images of the same scene need to be registered and fused.Multi-source remote sensing image fusion utilizes the specific techniques to remove redundant information from images of different sources in the same scene,combing with complementary information from them to generate a clearer,more accurate and more comprehensive image of the scene.Remote sensing image fusion has a wide range of applications in military and civil fields.In this paper,we focus on four kinds of heterogeneous source image fusion techniques.The main works are as follows:Firstly,a remote sensing image registration method based on improved speeded up robust features(SURF)in wavelet domain is studied,which lays the foundation for the following image fusion.When image registration is performed in the wavelet domain,the search range of the feature points can be reduced,and the time of feature points extraction and subsequent registration can be greatly reduced.At first,the reference image and the image to be registered were decomposed into the low-frequency components and high-frequency components by the wavelet transform,respectively.Then the lowfrequency component served as the input image of improved SURF method proposed in this paper,thus the coarse registration point pairs can be obtained: the dimension of descriptors was reduced by the principal component analysis method,the coarse registration of feature points were achieved by the criterion of bi-direction registration.Next,precise registration point pairs were gradually screened out by using the random sample consensus(RANSAC)method with the different distance thresholds twice.Finally the registration image was obtained.Experimental results show that the proposed method greatly improves the registration speed with the higher registration accuracy compared with the four algorithms such as the scale invariant feature transform(SIFT)method,the SURF method,the multi-scale registration method in wavelet domain using SURF,the method based on nonsubsampled contourlet transform(NSCT)and SURF.Then,a method for fusion of multi-spectral image and panchromatic image based on chaos artificial bee colony optimization and improved pulse coupled neural network(PCNN)in nonsubsampled shearlet transform(NSST)domain is proposed.Firstly,intensity hue saturation(IHS)transform is performed on the multi-spectral image.The histogram of panchromatic image is matched to the histogram of intensity component of the multi-spectral image.Then the intensity component of multi-spectral image and the new panchromatic image are decomposed by NSST,respectively.Next,the low frequency component is fused with the improved weighted fusion method.The mutual information is used as the fitness function and the optimal weighted coefficient is found by the chaos artificial bee colony optimization method,while the improved PCNN method is adopted for the fusion of high frequency components.Finally,the fused image is obtained by inverse NSST and inverse IHS transform.The experimental results demonstrate that the proposed method can effectively preserve spectral information,while improving spatial resolution of the fused image,and outperforms five fusion methods such as the IHS method,the method of non-subsampled contourlet transform(NSCT)combined with non-negative matrix factorization(NMF)and the method of NSCT combined with PCNN in the objective quantitative evaluation indexes such as information entropy and spectral distortion.And then,an infrared image and visible image fusion method based on target extraction and guidance filtering enhancement is proposed.Firstly,the two-dimensional Tsallis entropy and graphbased visual saliency(GBVS)model are used to extract the target region of infrared image.The visible image and infrared image are decomposed by NSST,respectively.The low frequency components of the visible image and infrared image are enhanced with guided filtering,respectively.The substitution method is employed to fuse the low frequency components of the visible image and the low frequency component of the infrared image,the high frequency coefficents adopt the maximum criterion of the sum of the coefficents absolute value of the same scale as the fusion rule.Finally the fused image is obtained by inverse NSST transform.Experimental results demonstrate that the proposed method can effectively highlight the target and preserve spectral information,while improving spatial resolution of the fused image,and is superior to five fusion methods such as the method based on the laplacian pyramid(LP)transform,the method based on wavelet transform(WT),the method based on stationary wavelet transform(SWT),the method based on non-subsampled contourlet transform(NSCT),the method based on target extraction and NSCT transform in the quantitative evaluation indexes.Subsequently,a fusion method of visible image and SAR image in Curvelet domain based on region characteristic and texture enhancement is discussed.The visible image and SAR image are decomposed by Curvelet transform,respectively,and the adaptive fusion rule based on improved regional variance is adopted in the low frequency component to avoid the loss of the target information in SAR image.Since the texture features are concentrated in the high frequency components,the improved local binary pattern(LBP)operator is used to extract the texture features from the high frequency components of the SAR image.For the fusion of high-frequency components,high-frequency fusion components are obtained by weighted fusion rules based on regional spatial frequency.Then,the texture features extracted by the improved LBP are added to the high-frequency fusion components to improve the contrast of the fused image.Finally the fused image is obtained by the inverse NSST transform.Compared with the fusion method based on NSCT,the fusion method based on multi-scale Top-Hat transform proposed in recent years and other three classical fusion methods,the proposed method can preserve and enhance the target and texture features of the images to be fused,and the contrast and the resolution of the fusion image are higher.Finally,a fusion method of infrared image and SAR image in complex contourlet domain based on joint sparse representation is proposed.The complex contourlet decomposition of the infrared image and the SAR image is carried out,respectively.The K-SVD method is employed to obtain the overcomplete dictionary of the low-frequency components of the two source images.According to the joint sparse representation model,the joint dictionary is generated.The sparse representation coefficients of the low-frequency components of the source image under the joint dictionary are obtained by the OMP method,and the sparse representation coefficients of the two low-frequency components are selected by the selection maximization strategy.The high frequency components are fused by two criteria of visual sensitivity coefficient and energy matching degree.Finally,the fusion image is obtained by the inverse complex contourlet transform.Compared with the fusion method based on NSCT,the fusion method based on sparse representation proposed in recent years and other three classical fusion methods,the proposed method can effectively highlight the salient features of the source images and inherit the information of the source images to the greatest extent.