Research on Geometric and Radiometric Correction Order and Its Impact for Remote Sensing Image

  • PhD Jiahui Xu, International Institute for Earth System Science, Nanjing University, China
  • Professor Zhaocong Wu, School of Remote Sensing and Information Engineering, Wuhan University, China

The correction of remote sensing image includes radiometric correction and geometric correction. Generally, radiometric correction is executed before geometric correction. Is this order a must? How much does the change of order influence on the radiance of pixel? And how much does it influence on quantitative remote sensing? Such kinds of research had been rarely reported. It is meaningful to quantitative analyze of the images reflectivity discrepancy caused by different radiometric and geometric correction orders. It will help to clear the influences caused by the changing correction orders, and also play a significant role on improving the accuracy of quantitative products. Additionally, the improvement of geometric correction methods brings higher efficiency and less dependence on ground parameters, which even makes calculating on orbit possible. But radiometric correction is still difficult to be performed by real-time because that the atmospheric environment parameters are hard to obtain on orbit. Therefore, the elasticity and instantaneity of image can be improved if doing geometric correction before radiometric correction.
In radiometric correction, both radiometric calibration and atmospheric correction recalculate the radiance of pixel. In geometric correction, coordinate transformation only repositions the pixel抯 geographical position without changing the radiance reflectance, while resample of geometric correction will recalculate the radiation value of pixel. Based on what mentioned above, this paper designs three processing flows with different correction orders to study the influence on pixel抯 radiance:
1.Radiometric calibration, atmospheric correction and geometric correction;
2.Geometric correction, radiometric calibration and atmospheric correction;
3.Radiometric calibration, geometric correction and atmospheric correction.
Then studies are carried out based on three resample methods including nearest neighbor, bilinear and cubic convolution respectively. In this study, three kinds of sensor images including ALOS, TM and CBERS-02B are chosen to do experiments, so there are 27 tests (3󫢫)to be performed. Through comparative analysis of the D-value calculated by retrieval images, we evaluate the discrepancy of image reflectance due to different correction orders.
Moreover, mathematical models to express the processing flow are built, from which we can analyze the causation of these discrepancies.
Conclusion is drawn from the experiment and mathematical analysis: when the nearest neighbor method is used to resample, the correction order does no impact on reflectance of image; when the bilinear method is used to resample, the change of reflectance due to correction order is small and not affect subsequent applications even it is ignored ; when cubic convolution is used to resample, the radiance changes due to correction orders is relatively larger, which will affect subsequent applications.
According to the mathematical modeling analysis we come to know the main reasons that cause this discrepancy are: radiometric correction is based on the radiance, while geometric correction is based on the digital number, and digital number is obtained from the quantification of radiance, so it is not a real physical quantity. The corresponding relationship between digital number level and radiance will be destroyed by geometric correction resample process, and the influence differs depending on the resample method, and the effect become greater with the increase of pixel number involved in the resample process. According to the mathematical modeling analysis, the offset in radiometric calibration is the main cause, which will be amplified through accumulation with the number of pixel increases in the resample process.
Based on above experiment and analysis, for the process of remote sensing image correction, the discrepancy due to different correction orders mainly depends on the resample method in geometry correction. While the nearest neighbor and bilinear resample methods have no or small impact, but the cubic convolution resample method affect much more, where processing order should be considered carefully. However, the process orders will not be limited if the geometric correction is based on radiance to calculate, which can realize the direct calculation on the orbit and meet the elasticity and instantaneity of image processing.

Keywords: Remote Sensing Image, Geometric Correction, Radiometric Correction, Processing Order