Two Feature-Level Fusion Methods with Feature Scaling and Hashing for Multimodal Biometrics

Abstract

ABSTRACTThis paper presents a new feature-level information fusion mechanism based on shuffle coding, called shuffle coding-based feature-level fusion (SC-FLF), for personal authentication. Our approach (SC-FLF) aims at constructing an information fusion mechanism to integrate features from the same or different feature spaces in which the ranges of feature values from different traits differ largely. In this mechanism, the shuffle-coding operator includes dimension adjustment, feature standardization, and fusion coding. This paper addresses two distinct methods, such as feature scaling and hashing, to standardize the range of independent features of data. The shuffle encoder of the SC-FLF in Method 1 uses a feature scaling and the resulting binary code represents the distance between a set of normalized feature values with 2’s complement. On the other hand, in Method 2, the shuffle encoder of the SC-FLF with hashing uses a projection framework for maximizing the features on a hyperplane and then quantizes the hash values as a sequence of binary codes. A XOR operation works as the fusion coding to produce the resulting fusion code. Three different types of fusion are designed to evaluate the fusion performance. Experimental validation illustrates that the proposed fusion methods for combining features in multimodal biometrics advances the recognition performance significantly.