Automatic features reduction procedures in palm vein recognition

Feature or dimensionality reduction has become one of fundamental problem in the field of pattern recognition such as biometrics. Selecting the number of feature or dimension has become one challenge. Instead selecting number of feature manually, this work proposed a procedure for feature reduction by finding the correlation between recognition rates and number of features. The procedure started with collecting recognition rates from available classes against a number of features and then calculated some variables from the distribution to be used as anchors for estimating number of features in case there are new classes to be added. This study was applied on a palm vein biometrics system which used DCT and k-PCA as features extraction method. The results of the experiment showed that the procedure was able to achieve a number of features that have an average offset of less than 6 from those obtained from direct observation and an average error of 1.1% from the real recognition rates.