Fusion Method Of Remote Sensing Images Based On Multiscale Geometric Analysis

As an important branch of the image fusion, remote sensing image fusion mainly investigates how to integrate image information from the different remote sensing sensors, and generate new images in order to get more complete, objective and essential cognition for the same thing or target. In recent twenty years, it has been a very popular research area, and becomes efficient means to solve the large number data. Especially with the launch of high-resolution remote sensing satellite like IKONOS, Quick-Bird and SPOT-5, the resolution of remote sensing image is higher than before, and the application of the remote sensing images have been more useful. The purpose of image fusion is to utilize the complementary and redundant information in multi-source images, to synthesize information intelligently, and to produce more compete and more reliable description and decision than one image, in order that the fused image should be more suitable for human visual perception or computer further processing. In this paper, most of the research works focus on the fusion method between panchromatic (Pan) image and multispectral (MS) images. The following are the main contents:(1) The developing of history and research situation of remote sensing fusion are discussed, and many traditional remote sensing fusion algorithms are summarized and compared.(2) Image preprocessing technique is a very important stage before fusion, its error directly related to the quality of the fusion image, therefore, image preprocessing is the prerequisite and basis for remote sensing image fusion. This paper briefly summarizes the image preprocessing techniques and introduces the geometry correction, image registration and histogram matched. This paper introduces the subjective and objective evaluation criteria, especially on the objective evaluation criteria, the evaluation parameters are classified from three aspects which are the statistic characteristics of a single fusion image, the relationship between the fusion image and the standard reference image, and the relationship between the fusion image and the original images.(3)Wavelet analysis has multi-resolution characteristics, and has been widely used in the image fusion field. However, the traditional fusion methods based on real wavelet transform have shift change issues. Therefore, fusion method based on the dual-tree wavelet transform for this issue is proposed. The dual-tree wavelet transform can keep the approximate shift invariance and has well direction analysis ability, while only import limited data redundancy, and achieved good results in other areas. In this paper, the author makes use of the complementarity between multispectral images at rich spectrum information and panchromatic images at higher spatial resolution, researches the pixel-image fusion algorithms to retaining the original spectrum information and improved spatial resolution, and as well as researches the remote sensing image fusion methods based on dual-tree wavelet transform. Theoretical analysis and experimental results show that dual-tree wavelet transform is more suitable for remote sensing image fusion than real wavelet transform, and it can fully replace the real wavelet transform based fusion methods.(4) As a new multi-scale geometric analysis tools, Curvelet is more appropriate for analyzing the image edges such as curve or linear characteristics than wavelet transform, and it has better approximation precision and sparsity description. In this paper, the second generation Curvelet transform theory and property are researched, and the image fusion method based on the second generation Curvelet transform is proposed, the panchromatic image and multi-spectral images are decomposed using Curvelet transform, then the Curvelet coefficients are fused with the fusion regular in the corresponding scales, and finally the fused coefficients are reconstructed to obtain fusion images. Multi-focus images and remote sensing images are taken as experimental data, and the experimental results show the effectiveness of the method. Then, using the shift invariance of dual-tree wavelet transform and the "Anisotropy" characteristic of second generation Curvelet transform, the fusion method based on the dual-tree wavelet transform and second generation Curvelet transform is proposed. The experimental results show the effectiveness of this method, and show that the second generation Curvelet transform is a well multi-scale geometric analysis tools, the remote sensing image fusion method based on second generation Curvelet transform can obtain more better fusion effect.