Image Registration With Artificial Neural Networks Using Spatial and Frequency Features

Image registration has been widely used in many fields such as medical imaging, remote sensing, and computer vision. It transforms multiple images of the same subject that are taken either at different times or from different points of view to the same coordinate system. In this study, two novel neural network based approaches are investigated for the registration of magnetic resonance images (MRI). They combine features in spatial domain (with scale invariant feature transform) and frequency domain (with discrete cosine transform or discrete wavelet transform) together to provide more robust feature extraction methods for image registration. Besides, the learning ability and nonlinear mapping ability of artificial neural network provide a flexible and intelligent tool for data fusion on feature matching and parameter estimation. The performances of the proposed approaches are studied and compared with other methods via computer simulations.