Dense registration of push broom hyperspectral imageries

Push broom hyperspectral imaging is a process where a number of 1-D linear images are continuously sampled from a line-scan camera. These linear images need to be realigned to form a 2-D scene image using direct geo-referencing measurements such as GPS/INS. Such kind of geometric rectification is not exact, resulting in distortions in spatial coordinates. Common linear transformations are inadequate for the registration of such imageries. Nonlinear image registrations need to be exploited instead. In this paper, we propose an efficient coarse-to-fine solution for the registration of push broom hyperspectral imageries. In the coarse registration stage, Piece-wise Linear transformation is used to compensate for the large geometric differences between two hyperspectral imageries. The registration results are refined using a point-to-point dense mapping in the fine registration stage. Experimental results show that the proposed solution is feasible for the registration of push broom hyperspectral imageries with good registration accuracy.