Multiresolution Analysis of SAR Data
Abstract
The 'Multiresolution Analysis of SAR Data' program has supported research work in five areas: (1) Geometric hashing theory can now be viewed as a Bayesian approach to object recognition. False alarm rates can be greatly reduced by using certain enhancements and modifications developed under this project. (2) Geometric hashing algorithms now exist for the Connection Machine. Recognition of synthetically-produced dot arrays has been demonstrated using a model base of 1024 objects. The work represents a substantial advance over existing model-based vision capabilities. (3) Algorithms have been developed for determining the translation and rotation of a sensor given only the image flow field data. These are new algorithms, and are much more stable than existing computer vision algorithms for this task. The algorithms might provide independent verification of gyroscopic data, or might be used to compute relative motion with respect to a moving scene object, or may be useful for motion-based segmentation. (4) Our theories explaining the Dempster/ Shafer calculus and developing new uncertainty reasoning calculi have been extended, and presented at a conference and were incorporated into the Bayesian interpretation of geometric hashing (see (1) above). (5) 'Wavelet Slice Theorem' has been developed in several different versions, any of which yields an alternate approach to image formation. The result may well provide a more stable approach to image formation than the standard Fourier-based projection slide theorem, since interpolation of unknown spectra values is better-founded.... Geometric hashing, Uncertainty reasoning, Wavelets.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1993
- Accession Number
- ADA264727
Entities
People
- Robert Hummel
Organizations
- New York University