Feature Extraction Using the Hough Transform.
Abstract
This thesis applied the normal straight line parameterization of the Hough transform to a variety of images using the accumulator method. Simple inputs were used initially to illustrate the distortion characteristics of the Hough transform due to rotations, scales and translations of an input. Making use of work performed by D. Casasent and R. Krishnapuram of Carnegie-Mellon University, the Hough transform was then applied to segmented and edged doppler images. A distort-and-compare routine, which makes use of the Hough space distortion characteristics, was implemented in the Hough transform domain to estimate input space characteristics of an object. Next, the Hough transform, generated using Fourier transform techniques, was applied to some of the same inputs to demonstrate that the accumulator method is actually a discrete version of the continuous Hough transform. An unsuccessful attempt at implementing the continuous Hough transform using computer-generated interferograms was outlined. A method of implementing the continuous Hough transform and its inverse using phase filters was presented as a suggestion for further research. Keywords; Theses; Pattern recognition; Computer generated holograms; Edges.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1987
- Accession Number
- ADA190679
Entities
People
- Marvin L. Hill
Organizations
- Air Force Institute of Technology