Optical Pattern Recognition of Partially Occluded Images

Chalasinska-Macukow, Katarzyna, Univ. of Warsaw

doi:10.1117/3.2265059.ch19

Book: International Trends in Applied Optics

Editor(s): Arthur H. Guenther

Author(s): Katarzyna Chalasinska-Macukow

Published: 2002

https://doi.org/10.1117/3.2265059.ch19

Author Affiliations +

Abstract

Optics encompasses a broad set of technologies and techniques for exploiting the properties of light. Many of the devices and systems of photonics are based largely on classical optical principles that could be fully exploited only after the discovery of new materials and the invention of efficient processes for bringing these new developments to the factory and the marketplace. In recent years, both material science and modern optics have developed in an atmosphere of interdisciplinary research and engineering, and together they have rapidly changed our modern world. The appearance of high-resolution fast spatial light modulators, CMOS cameras, and cheap laser diodes has enabled us to realize new photonic devices and introduces the possibility for new solutions. Image processing is a well-suited discipline for optical implementation because of the inherent 2D parallelism that optical processors can provide. Furthermore, optical data storage offers high storage capacity and fast parallel access to the stored information. The combination of optical storage (i.e., holographic memory) and optical pattern recognition (i.e., optical processor based on correlator architecture) can create an appropriate system of machine vision that is capable of real-time vehicle navigation and target acquisition, automated personal identification systems, or various content-addressable memory structures. The algorithms used to solve real recognition problems are usually time-consuming. For example, a machine that uses visual sensing for real-time vehicle navigation has access to a large amount of data about the surrounding environment, and extracting useful information from this large data set is not trivial. Personal identification based on the fingerprint pattern, the image of the face, eye retina, or iris structure, are examples of biometric recognition that require fast access to a large amount of data. A short time of access is also important in the case of content-addressable memory.