Real-time object recognition is, of course, one of the most important and attractive requirements in robot vision. Many methods have been developed for pattern recognition. Among these, correlation methods are of great importance. Shift and rotation invariances are the basic problem in recognition. A conventional matched spatial filter system is shift invariant but its shortcoming is that only objects in known orientations can be recognized.
The method presented in this paper preserves shift invariance while adding rotation invariance. Digital as well as optical implementations are presented.
KeywordsOptical System Input Image Inverse Fourier Transform Rotation Invariance Optical Processing
Unable to display preview. Download preview PDF.
- 1.Goodman J.W., Operations Achievable with Coherent Optical Information Processing Systems. Proc. IEEE 65 /1, 29–39, 1977Google Scholar
- 2.Hansen E.N., Goodman J.W., Optical Reconstruction from Projections via Circular Harmonic Expension, Opt.Comm. 24 /1, 269–272, (1978)Google Scholar
- 4.Macukow B., New Approach to Robotic Visual Processing, in: Theory and Practice of Robots and Manipulators, Proc. HO.A:an.Sy, Kogan Page, London eds. 74A.Morecki, G.Bianchi, K.Kgdzior. 1985, 255–259Google Scholar
- 5.Macukow B., Some Problem of Control of a Two-Link Anthoropomorphic.4anipulators. Proc. Symp.SyH000’85, IFAC, Barcelona, 467–469Google Scholar