Mutual Information-based Remote Sensing Image Nonlinear Registration

Non-linear registration of remote sensing images which is a tool for modifying thenon-linear aberrance of them always has a big influence to weather map which is theresult of image fusion. At present, researches of non-linear registration especially inthe field of remote sensing images is so little. After learning the some primary modelsof non-linear registration for the medical images and giving a full analysis ofcharacteristics of non-linear aberrance of the remote sensing images, we establish thepractical method of non-linear registration for the remote sensing images based onmutual imformation. This paper also make the experiment of the arithmetic, and thedetail of it is shown as below:At the first, we establish the registration method of delamination base on thecharacteristic of non-linear aberrance of the emote sensing images; secondlycalculate the barycenter, principal axis and some other data as the initial parametersof affine transform; the third, with the changing instance of mutual imformation, weuse the improved simplex optimization method which is syncretized with weightedcentroid method to finish the linear registration of small domain images; thefourth, find the controlled points set and linear transform matrix from affinetransform, and with the distortion coefficient, then could get the non-lineartransform matrix; finally, construct the non-linear mapping function with the lineartransform matrix and the non-linear transform matrix to carry out the non-linearinterpolation, we call it thin-plate spline interpolation, then could finish the non-linearregistration of the whole remote sensing images.The result of non-linear registration is fine according to the external evaluatingcoefficient: correlative coefficient, minimal share error, the signal to noise ratio. So it isshown that research of the non-linear registration of the remote sensing image in thispaper is preferable.