Radiometric Normalization of Multi-Temporal Images Using Kernel Canonical Correlation Analysis with Linear, Polynomial and Gaussian Kernels

Here a kernel method based on the established method known as kernel canonical correlation analysis (kCCA) is introduced to perform radiometric normalization of Chinese Gaofen1 (GF1) satellite images. It minimize image spectral differences between multi-temporal images without distinction of imaging conditions or the difference of reflectivity and perfectly eliminating the effects of nonlinear changes of features. We conduct radiometric normalization experiment with linear, polynomial and Gaussian (rbf) kernel functions to evaluate the performance from the characteristics of NIFs distribution and radiometric normalization results. The result of polynomial kernel has the highest similarity with the reference image, which means polynomial kernel is best suited for radiometric normalization.