Seminário do discente Pedro Henrique Lopes, dia 18 de julho às 13:30, na sala de seminários do DECOM

Seminário do discente Pedro Henrique Lopes, dia 18 de julho às 13:30, na sala de seminários do DECOM - ICEB III.

Título: Chimerical Dataset Creation Protocol Based on Doddington Zoo: A Biometric Application with Face, Eye, and ECG

Resumo: Multimodal systems are a workaround to enhance the robustness and effectiveness ofbiometric systems. A proper multimodal dataset is of the utmost importance to build such systems.The literature presents some multimodal datasets, although, to the best of our knowledge, thereare no previous studies combining face, iris/eye, and vital signals such as the Electrocardiogram(ECG). Moreover, there is no methodology to guide the construction and evaluation of a chimericdataset. Taking that fact into account, we propose to create a chimeric dataset from three modalitiesin this work: ECG, eye, and face. Based on theDoddington Zoocriteria, we also propose a generic andsystematic protocol imposing constraints for the creation of homogeneous chimeric individuals, whichallow us to perform a fair and reproducible benchmark. Moreover, we have proposed a multimodalapproach for these modalities based on state-of-the-art deep representations built by convolutionalneural networks. We conduct the experiments in the open-world verification mode and on twodifferent scenarios (intra-session and inter-session), using three modalities from two datasets: CYBHi(ECG) and FRGC (eye and face). Our multimodal approach achieves impressive decidability of7.20±0.18, yielding an almost perfect verification system (i.e., Equal Error Rate (EER) of0.20%±0.06)on the intra-session scenario with unknown data. On the inter-session scenario, we achieve adecidability of7.78±0.78and an EER of0.06%±0.06. In summary, these figures represent a gainof over28%in decidability and a reduction over11%of the EER on the intra-session scenario forunknown data compared to the best-known unimodal approach. Besides, we achieve an improvementgreater than 22% in decidability and an EER reduction over 6% in the inter-session scenario.