Research On The Multispectral Remote Sensing Image Fusion Technology

With the rapid development of sensor technology, image data with differentcharacteristics acquired by different kinds of remote sensors increased in largequantities, which brings a heavy burden to its storage, transmission and subsequentprocessing. Both complementation and redundancy usually exist among theinformation contained by the multi-source images. Therefore, image fusiontechnology received more and more attention, which focuses on extracting moreconcise and useful information with higher quality from source images. As a kind ofimportant remote sensing images, the research of multispectral remote sensing imagefusion technology has become the research emphasis of image fusion technology.And many excellent methods and algorithms have appeared in the past twenty years.Combining with the research work supported by National Natural ScienceFoundation of China and 973 Project, three types of multispectral remote sensingimage fusion algorithms are investigated in this dissertation, including multi-bandimage fusion algorithm, multispectral (MS) and panchromatic (Pan) image fusionalgorithm, multispectral and hyperspectral (HS) image fusion algorithm. Theresearch work and main results are listed as follows:(1) The AVA-RS algorithm based on orthogonal wavelet transform formulti-band image fusion After analyzing and investigating the relationship betweensource images, new measurements (Absolute Value Activity, AVA and RegionSimilarity, RS) are proposed to evaluate information quantity of an image orinformation redundancy between two images. And image fusion algorithm based onAVA and RS (AVA-RS algorithm) is proposed according to the properties ofcoefficients in orthogonal wavelet transform field. The algorithm separates spatialhigh- and low-frequency information by employing orthogonal wavelet transform,and the fusion rule is decided based on the values of AVA and RS measurements. Theexperimental results show that AVA-RS algorithm outperforms the congenericclassical algorithms in fused image quality and is suitable for the fusion of varioustypes of images, which indicates its good applicability.(2) The RWT-IOTF-SCC algorithm based on Integrated OrientationTexture Feature in redundant wavelet field and Significant Central CoefficientFor MS remote sensing image, texture feature is another important feature besidesspectral feature. And that is the reason why it should be given good consideration inMS image fusion. Based on texture energy method, the definition and calculationmethod of Integrated Orientation Texture Feature (IOTF) is constructed. And a newmulti-band image fusion algorithm (RWT-IOTF-SCC algorithm) is also proposed.The algorithm combines spectral features and morphological features in MS imageseffectively, preserves the spectral characteristics of source images and also takeseffect on fusion result of texture features into account. The experimental resultsshow that RWT-IOTF-SCC algorithm outperforms the congeneric classical algorithms for images with a large amount of texture features. The fused image canpreserve features (especially spectral features) of source images more effectively.(3) The X-LIE-LIEDP algorithm based on Local Information Entropy andits distribution property for MS and Pan image fusion Based on the analysis ofthe information quantity and its distribution property of effective spatial high-andlow-frequency information and spatial high-frequency noise in image, two newmeasurements named Local Information Entropy (LIE) and LIE DistributionParameter (LIEDP) are defined. And a new MS and Pan image fusion algorithm(X-LIE-LIEDP algorithm) based on LIE and LIEDP is also proposed. The algorithmis constructed on the foundation of HIS transform and multi-resolution analysis(MRA) combined image fusion model, and the fusion rule is decided according tothe characteristics of the LIE and LIEDP values. The experimental results show thatX-LIE-LIEDP algorithm is suitable for many types of MRA methods. In noise-freecondition, fused image quality of the algorithm is better than or equal to that of thecongeneric classical algorithms. In noisy condition, compared to the congenericclassical algorithms, the algorithm is capable of guaranteeing the spatial and spectralquality of the fused image, meanwhile, effectively resisting noise in source Panimage.(4) The LSRM-MPF algorithm based Local Spatial Recovery Model forMS and Pan image fusion By analyzing spatial relationship between remotesensing images of the same scene with different spatial resolutions, the definitionand construction method of Local Spatial Degradation/Recovery Model (LSD/RM)are proposed. And the LSRMs involved in MS and Pan image fusion are thoroughlydiscussed. Based on this, MS and Pan image fusion algorithm based on LSRM(LSRM-MPF algorithm) is proposed, whose main idea is to recover spatial featuresin source MS image according to LSRMs constructed between related images. Theexperimental results show that compared to the congeneric classical algorithms,LSRM-MPF algorithm is capable of achieving good fusion result with much lesscomputational cost. The controllability of the fused image quality, the equilibriumbetween the spatial and spectral quality of the fused image, and the reasonablecalculation cost make the algorithm more competitive.(5) The 3D-IDWT-MHF algorithm based on 3D Isotropic Discrete WaveletTransform for MS and HS image fusion After analyzing the 3D features of MS/HSimage, it is denoted that 3D method is more suitable for its analysis and processingthan 2D method. Based on analysis of properties of the coefficients of MS/HS imagein 3D Isotropic Discrete Wavelet Transform (3D-IDWT) field, MS and HS imagefusion algorithm based on 3D-IDWT (3D-IDWT-MHF algorithm) is proposed. Fourmajor steps constitute the algorithm: spectral resampling of MS image and spatialresampling of HS image, 3D isotropic discrete wavelet decomposition, waveletcoefficient combination and 3D isotropic discrete wavelet reconstruction. In addition,a new spectral resampling method based on ratio image (RIBSR method) and a newfusion rule (AS rule) for wavelet coefficient combination are also proposed. Thesimulative experimental results show that the RIBSR method can calculate the missing data effectively so that the quality of the fused image is guaranteed to somedegree. 3D-IDWT-MHF algorithm is capable of preserving both spatial and spectralfeatures of the source images, especially when AS fusion rule is employed. And itoutperforms fusion algorithms employing 2D analysis.(6) The HSDWT-MHF algorithm based on the mixed 3D discrete wavelettransform for MS and HS image fusion Based on the characteristics of spatial andspectral information in MS/HS image, a new MS and HS image fusion algorithm(HSDWT-MHF algorithm) is proposed. The algorithm employs the mixed 3Ddiscrete wavelet transform (HSDWT) as analysis method to separate high-andlow-frequency information in both spatial and spectral field. And the fusion rule isdecided according to the properties of the sub-block to which the wavelet coefficientbelongs. The simulative experimental results show that compared to classical fusionalgorithm using 2D discrete wavelet transform and 3D-IDWT-MHF algorithm(employing AS fusion rule), HSDWT-MHF algorithm is much more advantageous infused image quality enhancement. It is especially suitable for MS and HS imagefusion in which the spatial size of source MS image is much larger than the spectralsize of source HS image.