A non-parameterized method for co-registration of panchromatic and multispectral images

Abstract. Precise co-registration of panchromatic and multispectral images can be challenging due to the imperfect alignment of different sensors from the same platform or the involvement of different platforms. However, conventional methods heavily depending on the quality of feature matching and parameterized model fitting fail to yield an accurate result if the relative deformation between images is large. We propose a non-parameterized method that is free of such problems. A basic functional model is established with the consideration of equal radiance and smooth regularization. The local radiance deformation and the self-tuning weighting are then introduced to make the model more suitable for the specific requirement of co-registration. The model is finally solved with a twostage coarse-to-fine optimization approach. Our experiment on ZY-3 (China) images demonstrates its superiority over conventional methods, especially when large deformation due to terrain relief and sensor mis-alignment exists.