The Research And Realization Of Color Normalization And Cloud Removal In Large Scale Remote Sensing Image

Influenced by the differences of imaging time and environment,there are some problems in the mosaicked large scale high-resolution remote sensing images,such as color inconsistency and cloud interference.So far,a systematic scheme can normalize color,eliminate seam-lines and thin clouds haven’t been realized.This paper makes researches on automated color normalization for large scale mosaicked remote sensing image.Firstly,Wallis filtering and linear color transfer are introduced.Based on the idea of linear color matching,this paper improves finding optimal linear mapping by minimizing the variance difference between the image to be matched and the reference in SV and RGB color space.In order to extract the seam-line in the remote sensing image automatically,the research refers to edge features of haar-like and constructs features of sample can describe color mutation around the seam-line in HSV channels.The SVM classifier is applied to train seam-line detector.In order to realize the automatic color normalization of single large-scale remote sensing image,a patch-wise color matching algorithm is realized.The bilinear interpolation is also used to suppress the block artifacts caused by patch-wise matching.The patch-wise symmetric color matching method is proposed to further reduce the color cast around the stitching seam.Meanwhile,the proposed the local directional fusion method ensures color around the seam-line produces smooth transition.It calculates the color mapping of local area,then,weights by linear interpolation according to the distance between the pixels and the seam-line.The experimental results demonstrate that the these methods can effectively perform color normalization and eliminate stitching seam for remote sensing images of desert,mountainous and urban area.In order to eliminate the interference of thin clouds on the color normalization and highlight the ground objects,the patch-wise color matching is used to remove thin clouds as well.Compared with traditional thin cloud removal algorithms,the patch-wise color matching can suppress the thin cloud effectively.The color characteristics of terrains covered by thin clouds are closed to that of cloudless areas after patch-wise color matching.In conclusion,the paper provides a systematic scheme for automatic colornormalization of single large-scale mosaicked remote sensing image,which can equalize the color differences within the image,make the color transition naturally around the seam-line and reduce the sense of thin clouds.