Quantum-Limited Image Recognition
Abstract
Correlation-based methods for automatic image recognition are implemented using a position-sensitive, photon-counting detection system. It is demonstrated that the information provided by a small number of detected photoevents can be used to accurately estimate the cross correlation between a classical-intensity input scene and a reference (or filter) function stored in computer memory. A theoretical formalism is developed that describes the behavior of the quantum-limited correlation signal for complex filter functions. The theoretical predictions are verified experimentally using a position- sensitive photon-counting detection system. The speed at which the detection system operates makes this an effective technique for implementing correlation based methods for image recognition in real time, even when there is an abundance of input illumination. The estimation of moment invariants for image recognition is also considered. Experiments are performed that demonstrate that the information provided by a few thousand detected photoevents is sufficient to estimate moment invariants that remain unchanged when segmented input images are scaled, change in position, or undergo inplane rotations. The automatic recognition of images from within a cluttered environment is considered. The photon-counting detection system is used to implement a two-stage template matching algorithm to locate objects of interest from within cluttered scenes. Both two-stage matched filtering, and two-stage rotation-invariant filtering is considered. Theses.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1989
- Accession Number
- ADA218425
Entities
People
- Thomas A. Isberg
Organizations
- University of Rochester