A Novel Hybrid Multi-resolution Image Registration Algorithm with Curvelet Trasnform and Artificial Neural Networks

The purpose of image registration is to transform multiple images of the same subject taken from different points of view, times, depths, or sensors into one coordinate system. In this study, a novel hybrid image registration approach is developed based on curvet transform, discrete cosine transform, and artificial neural networks. Curvelet transform is an emerging multi-resolution analysis method that can effectively represent objects with highly anisotropic elements such as lines and curves. The proposed algorithm combines the orientation selective property of curvelet transform and the “energy compaction” property of discrete cosine transform together to provide a more efficient way to extract features of curvilinear structures from images. 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 on medical magnetic resonance images (MRI) via computer simulations.