Dimensionality Reduction Method Apply for Multi-view Multimodal Person Identification
Nassima Bousahba, Réda Adjoudj, Souaad Belhia, Lamia CHACHOU · International Journal of Computing and Digital Systems · 2022
In biometric systems, reducing the data dimensionality without compromising intrinsic information is essential in pre-processing high-dimensional data.Many states of the art use techniques to minimize the dimensionality of such data and avoid the so-called curse of dimensionality.When operating on limited datasets, supervised methods suffer from over fitting.Reducing the semi-supervised dimensionality in the next comparison or classification module can affect the recognition efficiency.This article introduces a novel multiview multimodal semi-supervised dimensionality reduction methodology that applies Multi-view Multidimensional scaling dimensionality reduction based on Gabor 2D-Log extraction features and Fuzzy Multiclass SVM classification (FMSVM), respectively.In addition, it examines its application to multi-view multimodal biometric processing, especially multi-view faces, and fingerprints.An experimental study was conducted, and the results emphasize that this methodology surpasses baseline supervised and semi-supervised methods.