In optical implementation of statistical image recognition, new optical transforms on large images for real-time recognition are of special interest. Several important linear transformations frequently used in statistical pattern recognition have now been optically implemented, including the Karhunen-Loeve transform (KLT), the Fukunaga-Koontz transform (FKT) and the least-squares linear mapping technique (LSLMT).1-3 The KLT performs principle components analysis on one class of patterns for feature extraction. The FKT performs feature extraction for separating two classes of patterns. The LSLMT separates multiple classes of patterns by maximizing the interclass differences and minimizing the intraclass variations.
Sing H. Lee,
"New Optical Transforms For Statistical Image Recognition", Proc. SPIE 0388, Advances in Optical Information Processing I, (23 December 1983); doi: 10.1117/12.935002; https://doi.org/10.1117/12.935002