Robust image registration under spatially non-uniform brightness changes

Based on a non-linear image representation, we propose a novel robust approach for registration of images under spatial brightness changes. The image registration is formulated as a two-stage hybrid framework combining both a new point-based algorithm and a robust estimation with M-estimators in a coarse-to-fine manner. With the point-based algorithm applied at the highest level of decomposition, the initial affine parametric model can be first estimated. Subsequently, the robust estimation using M-estimators is incorporated into the hierarchical framework for completeness. Experimental results demonstrate that our proposed algorithm achieves higher accuracy and efficiency than the approach by brightness variation modeling with low-order polynomial functions (BVM).