Robust multi-sensor image registration by enhancement of statistical correlation

Abstract

Image registration is a process to establish the spatial correspondence between images of the same scene, which are acquired at different viewpoints, at different times, or by different sensors. This paper deals with robust registration of the images acquired by two different sensors, namely, the electro-optic (EO) and infrared (IR) ones. The two approaches, feature-based and intensity-based image registration, are general for image registration. In the former approach, the selection of accurate common features is crucial for the performance, but features in an EO image are often different from those in the corresponding IR image. Hence, this approach is not adequate to register a pair of EO/IR images. In the latter approach, normalized mutual information (NMI) has been widely used as a similarity measure due to its high accuracy and robustness. NMI-based image registration methods utilize the statistical correlation between two images on the assumption that the correlation is global. However, it is often found that, in some areas of EO and IR images, statistical correlation is not high enough for robust registration. In this paper, we propose the two preprocessing schemes for improving the performance of the NMI-based registration. Both schemes try to enhance the statistical correlation between EO/IR images for fast and accurate registration. The first scheme, ESCR (extraction of statistically correlated regions), extracts the regions that are highly correlated to their corresponding regions in the other image. This extraction procedure is performed for each image, and the commonly extracted regions are used for NMI. The other scheme, ESCF (enhancement of statistical correlation by filtering), adaptively filters out two images to enhance statistical correlation between them. The proposed schemes are applied to NMI-based registration and the results are prospective for various pairs of EO/IR sensor images in terms of registration accuracy, robustness, and speed.