Stitching Methods For Satellite Images From Multiple TDI-CCD

The technology of multiple TDI-CCDs stitching is widely used in very high resolution space cameras for reaching a sufficient swath width and good imaging quality.Those multiple TDI-CCDs are usually non-collinear arranged and the sub-images are delivered as separated image files for raw data.As a consequence,those sub-images need to be stitched to obtain the larger image.This paper studies the highly accuracy stitching algorithms of 8 non-collinear sub-images of high resolution camera onboard a domestic Mapping Satellite.The main content and innovation points of the research are introduced as follows:1.Discuss imaging characteristics of the multiple non-collinear TDI-CCDs.Summarize some key factors that affecting the displacement in and across flight direction,including the location of different CCD chips,the drift angle controlling,the variation of integration time and topographic relief.Theoretical support and priori knowledge can be obtained through those analyses.2.Image space stitching algorithms based on image matching are systematically studied.The images generated by the individual TDI-CCD are simultaneous,geographical homologous and overlap slightly and so they can be stitched by tie points which are selected based on image matching.This paper studies the reliable extraction of tie points and key technologies include homonymy point prediction based on prior knowledge,bilateral matching,gross error elimination technique based on topographic and so on.What's more,different matching strategies are discussed here and some image space matching algorithms are designed.Two algorithms meet the requirement of a sub-pixel processing level.One is piecewise affine transformation and the other is single affine based on equalization of integration time.3.Object space stitching algorithm based on image sensor model is systematically studied.Firstly,this paper establishes the rigorous geometry model of each TDI-CCD chips according to the laboratory calibration parameters.In the second place,an inflight calibration approach of so called EXTERIO R-INNER way is presented,which,firstly,conduct exterior calibration to compensate the constant angular error,and then it introduces the additional polynomial coefficients based on self-calibration model to relive the internal CCD distortion.Finally,after the geometry model refinement,an object space stitching algorithm of so called VIRTUAL LONG CCD based on the terrain continuity is proposed and experiments show that it also reaches the sub-pixel level.All algorithms in this paper are designed based on the high resolution camera onboard a domestic Mapping Satellite and they have been validated by true satellite data which makes it useful to some extent.