Based Multi-resolution Analysis Of Remote Sensing Image Fusion Algorithm

Remote sensing is one of a modern imaging technique that the aerospace satellite captures the reflective electromagnetic wave of the earth and produces image product. Based on the width of the electromagnetic spectrum sensors capture, they could be classified into two categories, the panchromatic sensor and multispectral sensor. Panchromatic sensor covers a wide electromagnetic spectrum and produces images with very high spatial resolution. Multispectral sensor concentrates on a specified electromagnetic spectrum and produces images with high spectral resolution but low spatial resolution.For fully utilizing the complementary spatial and spectral characteristics of panchromatic and multispectral images, image fusion is proposed and considered to be a promising technique. The goal is to produce a data product that has high spatial resolution alike the panchromatic image and the spectral information the same as the multispectral image.Remote sensed image fusion technique, from early proposed intensity hue saturation (IHS) transformation, principal component analysis (PCA) based algorithms, to recently widely used wavelet based algorithms, the quality of the fused images are greatly improved, while the spectral distortion is reduced to a new low level. Nevertheless, recently used wavelet based fusion algorithms, because of its shift-invariant and lack of directionality properties, will introduce spectral distortion.First, after carefully analyzing the drawbacks of recently used wavelet based algorithms, we propose a Non-subsampled Contourlet Transform (NSCT) based remote sensed image fusion method. The algorithm introduces the NSCT analysis, while the Gibbs-like problem caused by shift-invariant is solved. Besides, NSCT analysis has nicer directionality, so it could easily extracts the spatial information from panchromatic image and injects it into multispectral images.Second, a Double Density Complex Wavelet Transform (DDCWT) based remote sensed image fusion approach is proposed. It down-samples the input image, but has the property of nearly shift-invariant, then greatly reduces the data redundancy. Thanks to the loose filter design condition, computation requirement has become a trivial problem. So it is very useful in situations that mass data demands processing.Finally, fusion strategy based on multi-resolution analysis is carefully investigated, and a fusion strategy based on whole spectral information is proposed. The algorithm classifies the whole scene into different spectral types. Then, in each type we calculate the statistical parameters as the weight coefficients during fusion. The algorithms avoid the instable and uncertainty properties of the weight coefficients calculated out by local window based approaches.