Remote sensing image registration based on the HEIV model

Abstract

In Earth observation missions, a large number of remote sensing images (RSIs) are obtained through satellites and aerial photography. RSIs are often huge and they generally have large edge distortions. Therefore, registration is required when the RSIs are used for mapping and change-detection. Since the RSIs usually have complex variation and the large distortions, the current existing global image registration algorithms cannot achieve high-precision matching of the target area. We propose a method based on the Heteroscedastic error-in-variables (HEIV) model to solve the registration problem of RSIs. Though HEIV model, the registration parameter estimation problem can be expressed as a quasi-linear form of the design matrix, which consists of observation vectors and parameter vectors with heteroscedasticity. Based on the image pre-processing and feature intensity, the HEIV model is utilized to estimate the registration parameters. Experiments are carried by using the RSIs taken with helicopter in comparison with the method based on Harris corner registration and the method based on SIFT registration, respectively. Compared with the Harris, the best group’s accuracy is increased by 54.76%. In comparison with the SIFT algorithm, the T-Rank group’s increment on RMSE is 1.88%. The results show that our method has more advantages in the parameter estimation of RSI registration and it can be better applied to Earth observation missions.