Relative radiometric correction of high-resolution remote sensing images based on feature category

The assumption that the spectral responses of different types ground objects in different periods have the same linear relationship in traditional relative radiometric correction (RRC) is insufficient for the analysis of high-resolution remote sensing images. For this reason, improvement was made based on PIF method, and a new RRC method for high-resolution remote sensing images considering ground object classes was proposed. First, histogram of oriented gradient feature was adopted to select unchanged regions. Then, PIF points were further selected from the unchanged regions using correlation coefficients. Combining with bands and object classification results, the selected PIF points were divided into groups. Finally, through least square regression analysis, the gain and offset were obtained, and the images to be corrected were corrected according to bands and ground object classes, and combined into the corrected images. The new RRC method experiments on Geoeye-1 and Ikonos high-resolution images of Urumqi City in Xinjiang Province showed that the proposed method performs better, with better visual effect and smaller root mean square error than the existing RRC methods.