Using Linear Discriminant Analysis to Fuse Bimodal Biometrics Traits in Complex Space

In this work, we propose a simple and efficient approach to using linear discriminant analysis (LDA) to fuse the traits of bimodal biometrics in complex space. The proposed approach consists of two phases, the phase of obtaining transform axes (the first phase) and the phase of feature calculation (the second phase). The first phase calculates transform axes using a linear discriminant analysis (LDA) scheme on each of the two traits of bimodal biometrics, respectively. The second phase combines the two kinds of transform axes to new transform axes in the form of complex vectors and denotes each sample by a complex vector. The proposed approach extracts features from the sample by projecting the complex vector to denote the sample of bimodal biometrics onto the complex transform axes. The proposed approach is able to convert more information of the sample than a conventional and direct feature level fusion approach. The results of extensive experiments show that the proposed approach can obtain a higher accuracy than previously complex-vector or matrix based feature extraction approaches.