Heterogeneous Face Recognition attempts to generate face image cropped over multiple techniques, such as visible, near-infrared and existing HFR techniques often rely on incomplete models or features artificially incapable of fully accounting for complex cross-domain differences Using a densely connected network structure called Convolutional Neural Networks (CNNs) and DenseNet to improve feature propagation and reuse, The study presents a deep learning for HFR in this study. A DenseNet-based model is build that can train robust and fully discriminative features for HFR. The study also develops a unique loss function that combines cross-entropy and center loss mixing to improve intraclass compression and interclass separation of learned features The study demonstrates that the proposed technique outperforms previous deep learning algorithms for HFR, 2005. state-of-the-art performance and achieves. The study also investigate how network configurations and loss functions affect the efficiency of the HFR. Proposed findings support the utility and superiority of our approach to HFR.
Advancing Heterogeneous Face Recognition Through Convolutional Neural Networks DenseNet
2023-11-22
489674 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Convolutional Neural Networks for Traffic Sign Recognition
TIBKAT | 2021
|British Library Online Contents | 2018
|Human Face Recognition using Neural Networks
British Library Conference Proceedings | 1994
|