Local cross-modality image alignment using unsupervised learning
We propose a method for automatically aligning images with local distortions from different sensors, using real images instead of calibration objects. The algorithm has three components. First, we extract intensity discontinuities, because this is a feature that is likely to show up across modalities. Second, we use a correlation scheme that averages over time rather than space, for high precision. Third, we propose an architecture and a learning scheme that learn the correlation surfaces over time and implement the image coordinate transform.
- Barniv, Y. and Casasent, D. (1981) Multisensor image registration: experimental verification, SPIE 292: Processing of images and data from optical sensors, 160–171.Google Scholar
- Knudsen, E.I. (1983) Early auditory experience aligns the auditory map of space in the optic tectum of the barn owl, Science 222, 939–942.Google Scholar
- Kohonen, T. (1984) Self-organization and associative memory, Springer-Verlag, Berlin.Google Scholar
- Pearson, J.C., Sullivan, W.E., Gelfand, J.J., Peterson, R.M. (1987) A computational map approach to sensory fusion, AOG/AAAIC Proc. Joint conf. on merging tomorrow's technologies with defense readiness requirementsGoogle Scholar