Pattern recognition with fast feature extraction
М.Г. Находкін, Yurij S. Musatenko, Vitalij N. Kurashov · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 1998
The comparison of recently suggested algorithms of fast approximate Karhunen-Loeve (KL) transform when applied to pattern recognition problem is presented. It is known that adequate signs selection is still the problem. In the paper we consider several algorithms that can perform fast and qualitative signs selection. Among them are wavelet based algorithm of approximate KL transform, 2D algorithm of KL transform, algorithm with projection into proper orthogonalized basis, and real time algorithm of KL transform. To compare quality of the algorithms they are applied to human faces recognition problem. The efficiency of the all abovementioned algorithms is demonstrated.