An object-oriented relative radiometric normalization method using high resolution remote sensing images

The assumption that the spectral responses of different types of objects in different periods have the same linear relationship in traditional relative radiometric normalization is insufficient for the analysis of high resolution remote sensing images. Object-oriented relative radiometric normalization for high resolution remote sensing image change detection is proposed in the paper based on the assumption that the spectral responses of different types of objects in different periods have different linear relationships. Firstly, image objects are divided into two categories: changed and unchanged by correlation coefficients. Secondly, gains and offset parameters are calculated by an analysis of a random sampling consensus based on unchanged image objects. Thirdly, gains and offset parameters of the unchanged image objects which are most similar with the changed image objects are assigned to the changed image objects. Lastly, the image objects are corrected using gain and offset parameters. Experiments on high resolution remote sensing images verify the effectiveness of the proposed method.