With the rapid development of economy, China’s urbanization steps into the fast lane gradually. In order to provide effective supervision and administration for the development of urban, it is urgent to extract information from massive amounts of urban data which is helpful to decision making and intellectual support. As the main form of urban data, image is the most important component. The most useful role of urban image is to act as a carrier of information, which makes it possible to conduct feature extraction, classification and interpretation. However, to extract ground features from urban images accurately and efficiently is the prerequisite for image interpretation. Image enhancement, image segmentation, image recognition and the image compression are all image processing methods in urban fields, besides, image fusion is important too. The key technologies of fusion are fully realize the properties of objects, suitable select the data according to spectrum differences between various remote sensing data, right image correction, precise image registration and other methods. According to the researches, we should pay attention to the following points in the process of the acquisition, pre-processing, fusion, evaluation and interpretation for urban images.Although urban image fusion was widely concerned, it is not a independent discipline or branch, in most situation it is only the auxiliary tool for interpretation process. And there was not any professional research field between remote sensing interpretation and earth physical disciplines, and not specific literatures that focus on a large amount of urban remote sensing data and the related application in image fusion. Wavelet transform, as an excellent multi-resolution analysis method, has many advantages, but it is shift-variant, can only captures limited directional information and can not optimally represents high-dimensional functions that contain curves and singularities in high dimensional functions. As a result, the curve can not be reconstructed accurately and singular lines or surfaces can only be approximated by singular points, so the contours and textures of the image are blurred. Meanwhile, square shaped wavelet basis can not make full use of geometric regularity of image, and is not the most sparsely function. The endless image types lead to a variety of image characteristics, what kind of fusion algorithm is suitable for what kind of image is not fully discussed yet, and what kind of fusion framework should be take in specific situation is not exactly concluded. As for image evaluation, all sorts of subjective and objective indicators do not form a unified system, and the selection of indicators is relatively subjective.According to the above mentioned facts, the article puts forward some suggestions, and the main work covered in the article can be summarized as follows:(1)The research background and significance of multi-source image fusion technology are described, and the urban remote sensing data and its application in a number of social fields are detailed discussed. The working mechanism, structure and working procedure for three levels image fusion algorithms are analyzed. The standard evaluation system for image quality is studied, by studying on CC(correlation coefficient) and SAM(spectral angle mapping), a new metric ISAM(improved spectral angle mapping) is proposed to evaluate the similarity of remote sensing image, which is more sensitive to the differences between two images.(2)The acquisition mode and sensor types of urban remote sensing images are briefly introduced, the working principles and characteristics of main remote sensing image fusion methods are adequately discussed. The characteristics and development trend of remote sensing earth observation technology are analyzed, two supervised classification are introduced and an inhabited area are automatically interpreted by taking texture as an example. The mutual relation between the geophysics disciplines and remote sensing image interpretation are studied in aspects of physics, geological depth and time.(3)Analyze the working mechanism and characteristics of wavelet transform and point out its shortcomings in reconstruction process for edges and contours according to the latest progress of the multi-scale analysis theory. Introduce a more perfect multi-scale transform-- non-subsamped contourlet transform(NSCT) theory, which is characterized by multi-scale, multi-direction, anisotropy and shift-invariant. By describing the overall structure, non-subsampled pyramid filter bank and non-subsampled directional filter bank of NSCT, discuss the feasibility of image decomposition, reconstruction and fusion techniques in NSCT domain. And establish the image fusion framework in NSCT domain, and finally the high and low frequency sub-images are introduced.(4)Discuss the feasibility of introducing non-negative matrix factorization(NMF) into the field of image fusion by studying the concept, principle and algorithm classification. Due to the non-negative constraints imposed on matrix elements, the generated base matrix has a specific physical meaning, which making it has a wide range of applications in the field of pattern recognition, target detection, medicine, biomaterials and image fusion. By studying the basic principle of Lee-Seung NMF(LSNMF) algorithm, a modified NMF(ANMF) algorithm is proposed and described. The fusion experiments show that the overall performance of ANMF is better than that of LSNMF, but ANMF is required to recalculate the parameters in each iteration step which makes it run longer than LSNMF.(5)Denoising methods in spatial domain and frequency domain are introduced respectively, the results reveal that the signal to noise ratio(SNR) of transform domain are superior to that of spatial domain. Analyze the feasibility of PCNN performed in image fusion by introducing the model and operation characteristics of pulse coupled neural networks(PCNN) in detail. Proposed an improved PCNN image fusion method in NSCT domain-- NSCT + PCNN, and put it in use of urban image fusion, as a result, the definition or activity level is increased. The ANMF theory is applied in NSCT domain to form NSCT + ANMF fusion method and is performed in urban image fusion. Finally, by comparison with NSCT + PCNN, NSCT + ANMF, basic NSCT and wavelet image fusion methods, the experimental results verify the effectiveness of various NSCT fusion methods.By conducting the paper, one can learn about the urban remote sensing data and its usage in different application areas. The proposed ISAM index plays a positive role in promoting objective quality evaluation system. Discussions on relation between disciplines of geophysics and remote sensing imagery interpretation deepen ones understanding about mutual correlation of spatial depth, geological rules and phenomena. Studies on NSCT, NMF theories and corresponding algorithms in contourlet domain verified the effectiveness of algorithms in urban image fusion, and try to make useful supplement and description in urban image fusion. |