15 April 1983 Linear Invariant Multiclass Component Spaces For Optical Pattern Recognition
Author Affiliations +
Proceedings Volume 0422, 10th Intl Optical Computing Conf; (1983) https://doi.org/10.1117/12.936139
Event: 10th International Optical Computing Conference, 1983, Cambridge, United States
Optical processing systems which perform linear transformations on image data at high rates are ideal for image pattern recognition systems. As a result of this processing capability, the linear opera-tion of matched spatial filtering has been explored extensively for pattern recognition. For many practical pattern recognition problems, however, multiclass filtering must be used to overcome the variations of input objects due to image scale changes, image rotations, object aspect differences and sensor differences. Hester and Casasent have shown that a linear mapping can be constructed which images all the class elements of a multiclass set into one out-put element or value. This special multi-class filter concept is extended in this paper to show that a subspace of the multi-class set exists that is invariant with respect to the multiclass mapping under linear operations. The concept of this in-variant space and its generation is detailed and a single example given. A typical optical processing architecture using these invariant elements as filters in an associative pattern recognition system is also presented.
© (1983) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Charles F. Hester, Charles F. Hester, } "Linear Invariant Multiclass Component Spaces For Optical Pattern Recognition", Proc. SPIE 0422, 10th Intl Optical Computing Conf, (15 April 1983); doi: 10.1117/12.936139; https://doi.org/10.1117/12.936139

Back to Top