Image alignment by parameter hypersurface learning
Hyun‐Chul Choi, B.H. Ahn · Electronics Letters · 2016
A new concept of image alignment method is introduced based on machine learning technique. The proposed method estimates image alignment parameters by using a regression hypersurface, which maps image feature space onto alignment parameter space. The regression hypersurface is obtained by learning an appropriate type of neural network from a large number of training samples, which consist of image features and corresponding alignment parameters. Compared with the previous method that utilises a regression hyperplane, this method achieved a significant improvement in convergence range to apply for alignment of rectangular region in long term image sequences in low frame rate.