Author/Editor		Cukjati, David; Pavešić, Nikola
Title		Kodiranje statistično neodvisnih značilk na področju razpoznavanja vzorcev
Translated title		Encoding of statistically independent features in pattern recognition
Type		članek
Source		In: Zajc B, Solina F, editors. Zbornik 5. elektrotehniške in računalniške konference ERK'96. Zvezek B. Računalništvo in informatika, umetna inteligenca, robotika, razpoznavanje vzorcev, biomedicinska tehnika, študentski članki; 1996 sep 19-21; Portorož. Ljubljana: Slovenska sekcija IEEE,
Publication year		1996
Volume		str. 267-70
Language		slo
Abstract		The paper deals with a technique for optimal source encoding of features in pattern recognition problems. Two different approaches are discussed. In the first we search for optimal number of threshold values, which do not detrimentally affect the probability of classification error if the Bayesian decision rule is used. The number of threshold values is equal to the number of separation values, which are the points of intersections of decision funcions. In the second approach the number of threshold values is fixed. We are searching for such threshold values for which the probability of classification error is minimal. If there is less threshold values than separation values, the probability of classification error is increased.
Descriptors		PATTERN RECOGNITION BAYES THEOREM