Use of Statistical Methods for Dimensionality Reduction in Hand Shape Identification Employing Radon Transform and Collaborative Representation Based Classification

Abstract

Hand shape based authentication has long been established as an effective method of biometric identification for access control and security. This paper presents the use of the different statistical measures like mean, median and standard deviation for feature dimensionality reduction following Radon transform along an optimal direction for each query image. Subsequently, the feature vector of the query image was coded over similarly processed training samples from all classes and the Regularized Least Square (RLS) method was employed to identify the query image as a member of the class which produces the least reconstruction residual. It was experimentally demonstrated that the overall performances of CRC based solutions were significantly better than that of artificial neural network (ANN) based classifiers, utilized for identical problems on the same database.