Feature selection for novel fingerprint dynamics biometric technique based on PCA

Fingerprint dynamics is a recently introduced behavioral biometric technique based on the time derived parameters from multi instance finger scan actions. Various related features can be extracted from recorded time stamps. However, not all of them contribute in improvement of classification accuracy and may result in high dimensionality of the data. High dimensionality leads to higher computation cost for calculating the features, and low classification rate. Thus it is crucial to select the best features for efficient system performance. Principal Component Analysis (PCA) is a popular technique for dimensionality reduction and has been applied to a wide number of applications. However conventional PCA based methods have a disadvantage of using all the features for transforming to lower dimensional space. In this paper, we follow a method based on PCA, which selects the most dominating features subset out of the feature pool at hand, without transforming the original features. The performance of selected features is assessed using various classification paradigms. The result ascertain successful selection of dominant feature subsets of fingerprint dynamics using PCA.