Improved Techniques for Relative Radiometric Normalization of High Resolution Satellite Images

  • Afify H


Relative radiometric normalization (RRN) is essential for change detection applications. Pseudo Invariant Features (PIF) and Dark and Bright set (DB) are widely used methods to perform RRN. The most important step in these methods is the identification of the normalization targets that should be used to derive the normalization coefficients (gain and offset). This paper first introduces the criteria formula to nominate the normalization targets of typical PIF and DB for QuickBird images. Then, some modifications are suggested to refine the selected normalization targets and to derive the normalization coefficients. The typical and modified PIF and DB methods are applied to normalize two pairs of QuickBird images covering parts of Tanta and Alexandria governorates in Egypt. Finally, the normalized images using the typical and modified PIF and DB methods are examined visually and compared statistically. It was found that the proposed modifications have significantly improved the accuracy of the normalized images.