Remote Sensing Image Fusion Method Based On Pixel-level

With the availability of multi-source data of remote sensing, it is convenient to receive multi-sensor and multi-scale images. These multi-source data have their own characteristic respectively. Images fusion is a technique that use the multi-source data at the same time, and provide fused images that increased interpretation capabilities and more reliable information.In this paper, the author focused on image fusion based on pixel level to improve spectral information fidelity, spatial information syncretizing and auto-classification precision of fused image. Based on analyses on current fusion methods, some basic study is performed with a comprehensive analysis and summarization of the former research work on image fusion. The main conclusions are as follows.(1)Based on analysis of spectral distortion of traditional IHS transform and spatial information syncretizing of traditional PCA transform, this paper discussed optimized approaches of traditional IHS transform with detailed. Modified Intensity-Hue-Saturation fusion method and Modified Principal Component Analysis fusion method are presented to overcome the spectural distortion of the traditional IHS transform and the shortcoming of the traditional PCA transform respectively. In a case of medium resolution image (Landsat 7 ETM+ image) and high resolution image (QuickBird image). The result shows that in terms of spatial information syncretizing, Spectral information fidelity and auto-classification precision, the MIHS transform and the MPCA transform get better result than traditional IHS transform and traditional MPCA transform respectively, and reduce the spectral information distortion of traditional IHS transform, and Enhances the spatial information syncretizing of traditional MPCA transform.(2)In order to overcome the shortcoming of SVR fusion method, a Spectral Range based Fusion Method is presented with a detailed analysis of principle of SVR fusion method. The SRFM transform is compared with SVR transform in a case of medium resolution image and high resolution image. The result shows that in terms of spatial information syncretizing, spectral information fidelity and auto-classification precision, the SRFM transform get better result than SVR transform. The parameter of the fusion model has obvious physical meaning and need not computed repeatedly for the same sensor.(3)In terms of visual effects, traditional fusion method based on Discrete Orthogonal Wavelet Transform gets unsatisfied result. Therefore, a new spatial adaptive method based on Discrete Stationary Wavelet Transform was promoted. The DSWT transform is compared with traditional DOWT transform in a case of medium resolution image and high resolution image. The result shows that in terms of spatial information syncretizing, spectral information fidelity and auto-classification precision, the DSWT transform get better result than DOWT transform.(4)The effectiveness of four new methods, MIHS,MTC,SRFM and DSWT, are assessed for spectral information fidelity, spatial information syncretizing and auto-classification precision respectively. The result shows that in terms of spatial information syncretizing, spectral information fidelity and auto-classification, DSWT transform get the best fusion result, followed by SRFM transform,MPCA transform and MIHS transform in both medium and high resolution image. But considering the practical application and the algorithmfusion efficiency, the SFRM transform is best, followed by MIHS transform, MPCA transform and DSWT transform respectively.