MOBIO D3.4: Description and evaluation of advanced algorithms for uni-modal authentication

speaker and face authentication, speech recognition, face detection

This document presented the advanced unimodal algorithms for face and speech authentication developed in the MOBIO project. The developed algorithms were compared against each other and against the baseline systems described in D3.2 [5].

This deliverable describes and evaluates the advanced unimodal systems for face detection, face point localisation, face veri cation and speech veri cation that were developed within the MOBIO project. In face detection, the presented advanced algorithms are based on using novel features and more ecient detection post-processing techniques. The proposed face point localisation method introduces a new local and global constraint model for the point locations. In face veri cation, advanced feature extraction and score normalization strategies are employed, whereas the advanced speech veri cation systems are based on using the joint factor analysis. All the developed algorithms are evaluated against the baseline systems described in project deliverable D3.2, showing clear improvement in veri cation performance both in the face and the speech modality.