Continuous Face Aging Generative Adversarial Networks

Jeon, Seogkyu; Lee, Pilhyeon; Hong, Kibeom; Byun, Hyeran

doi:10.1109/icassp39728.2021.9414429

Cited by 7 publications

(2 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…iris having high permanence and face having low permanence, are chosen to build unimodal deep learning models M1 and M2 from iris and face feature vectors fv1 and fv2 respectively. Face images are fed to the convolutional neural networks to fetch features and from Iris images, Kekre's median codebook [29] is generated to feed it to Long-short term memory for the respective classifier's construction. b.…”

Section: International Journal On Recent and Innovation Trends In Com...mentioning

confidence: 99%

Age-Adaptive Multimodal Biometric Authentication System with Blockchain-based Re-Enrollment

Sawant,

Bharadi,

Prasad

et al. 2023

IJRITCC

View full text Add to dashboard Cite

In the long run, a significant time gap between enrollment and probe image challenges the model's prediction ability when it has been trained on variant biometric traits. Since variant biometric traits change over time, it is sensible to construct a multimodal biometric authentication system that must include at least one invariant trait, such as the iris. The emergence of Deep learning has enabled developers to build classifiers on synthesized age-progressive images, particularly face images, to search for individuals who have been missing for many years, to avail a comprehensive portrayal of their appearance. However, in sensitive areas such as the military and banks, where security and confidentiality are of utmost importance, models should be built using real samples, and any variations in biometric traits should trigger an alert for the system and notify the subject about re-enrollment. This paper proposes an algorithm for age adaptation of biometric classifiers using multimodal channels which securely update the biometric traits while logging the transactions on the blockchain. It emphasizes confidence-score-based re-enrolment of individual subjects when the authenticator module becomes less effective with a particular subject's probe image. This reduces the time, cost, and memory involved in periodic re-enrolment of all subjects. The classifier deployed on the blockchain invokes appropriate smart contracts and completes this process securely.

show abstract

Section: International Journal On Recent and Innovation Trends In Com...mentioning

confidence: 99%

Age-Adaptive Multimodal Biometric Authentication System with Blockchain-based Re-Enrollment

Sawant,

Bharadi,

Prasad

et al. 2023

IJRITCC

View full text Add to dashboard Cite

show abstract

“…fm = fi + fa) is used to describe the mixed features and a decoupling method is proposed for decomposition. In [52], the mixture is modelled as the multiplication of the features (i.e. fm = fi × fa) and the decomposed element can be expressed as…”

Section: B Attention-based Feature Factorization Modulementioning

confidence: 99%

Multi-View HRRP Generation With Aspect-Directed Attention GAN

Song

Zhou

Yang

et al. 2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

In radar automatic target recognition (RATR), highresolution range profile (HRRP) has received intensive attention due to its low computational cost. As HRRP is sensitive to the aspect of the target, a training set covering sufficient aspects is essential to the success of a RATR model, which is however intractable in complex environment with noncooperative targets. In this paper, an aspect-directed attention generative adversarial network (ADA-GAN) is proposed to generate multi-view HRRPs using real samples from few aspects. The key is that the HRRPs from the similar targets share the same aspect variation pattern. Hence, an HRRP is decomposed into its identity and aspect features via an aspect-directed disentangled representation network with self-attention modules. In the training stage, the decomposition network and the aspect variation pattern are learned from full aspect samples of cooperative targets. When generation, the desired multi-view HRRPs of the noncooperative target are synthesized by its identity features extracted from few aspect samples and the learned aspect variation pattern. Three types of experiments on the simulated and measured datasets demonstrate the generation performances of our method. First, the generated HRRPs are visually compared with the truth. Second, the similarity of the scattering center power and handcrafted feature distributions are quantitatively evaluated. Finally, recognition experiments verify the feasibility of data augmentation with the generated HRRPs. Extensive results show the superior performance of our method over other state-of-theart methods.

show abstract