An improved Dark Object Subtraction (DOS) method was introduced for Landsat 8 multispectral satellite image in this paper. The main factors including Rayleigh scattering, Mie scattering in path radiance, as well as the other satellite image parameters (such as height modification, slope distance and azimuth), ware considered in the algorithm. The algorithm consists of three steps. First, starting band haze values are selected using histogram of a single image. Then predicted haze values were calculated using a known scattering model and the multispectral normalized gains and offset values. Finally, final predicted haze values are obtained by the predicted haze values and haze values. Compared with other improved Dark Object Subtraction methods, the result of this algorithm is more realistic on geographical object recognition on NDVI (Normalized Difference Vegetation Index) and NDWI (Normalized Difference Water Index) data.
In order to solve the problems that image’s entropy of information decline obviously and boundary line phenomenon appear obviously in the processing of aerial remote sensing image’s mosaic, an image mosaic approach is presented in this paper, which uses wavelet image fusion based on structure similarity and is capable of creating seamless mosaics in real-time. The approach consists of three steps. First, the overlapping area of two aerial images is extracted. Then, the two overlapping area images are fused adaptively by the method of multi-layer wavelet decomposition based on the structure similarity and appointed regulation. Finally, weighted average fusion is used again to avoid the visible boundary line for the both sides of the boundary of the above fusion image. Experimental results show the entropy of information, sharpness and standard deviation have been improved significantly, and the boundary line has been eliminated observably.