A preliminary investigation on the sensitivity of COTS face recognition systems to forensic analyst-style face processing for occlusions

Juefei-Xu, Felix; Pal, Dipan K.; Singh, Karanhaar; Savvides, Marios

doi:10.1109/cvprw.2015.7301316

Cited by 19 publications

(9 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Compared to other commonly used distance measurement such as ℓ 1 -norm, ℓ 2 -norm, NCD exhibits the best result [48][49][50][51][52][53][54][55]68,69]. The result of each algorithm is a similarity matrix whose entry SimM ij is the NCD between the feature vector of probe image i and target image j.…”

Section: Experimental Setup Overviewmentioning

confidence: 97%

Multi-class Fukunaga Koontz discriminant analysis for enhanced face recognition

Juefei-Xu

Savvides

2016

Pattern Recognition

Self Cite

View full text Add to dashboard Cite

Section: Experimental Setup Overviewmentioning

confidence: 97%

Multi-class Fukunaga Koontz discriminant analysis for enhanced face recognition

Juefei-Xu

Savvides

2016

Pattern Recognition

Self Cite

View full text Add to dashboard Cite

“…Metrics: In addition to the FID and Inception Score, we used metrics such as PSNR [16], SSIM, OpenFace [1] feature distance under normalized cosine similarity (NCS) [15] and PittPatt face matching score [17] to measure fidelity between the original and reconstructed face images. The last two are off-the-shelf face matchers that can be used to examine the similarity between pairs of face images.…”

Section: Evaluations On Face Completion Tasksmentioning

confidence: 99%

RankGAN: A Maximum Margin Ranking GAN for Generating Faces

Juefei-Xu

Dey

Boddeti

et al. 2019

Computer Vision – ACCV 2018

Self Cite

View full text Add to dashboard Cite

We present a new stage-wise learning paradigm for training generative adversarial networks (GANs). The goal of our work is to progressively strengthen the discriminator and thus, the generators, with each subsequent stage without changing the network architecture. We call this proposed method the RankGAN. We first propose a marginbased loss for the GAN discriminator. We then extend it to a marginbased ranking loss to train the multiple stages of RankGAN. We focus on face images from the CelebA dataset in our work and show visual as well as quantitative improvements in face generation and completion tasks over other GAN approaches, including WGAN and LSGAN.

show abstract

“…Juefie-Xu et al [41] conducted a study on forensic face processing techniques for occlusion and cropping. They used commercial offthe-shelf (COTS) components to build the face recognition system.…”

Section: Related Workmentioning

confidence: 99%

RDUP3: Relative Distance based User Profiling from Profile Picture in Multi-Social Networking

Vasanthakumar¹,

Shenoy²,

Venugopal³

2017

IJCA

View full text Add to dashboard Cite

User Profiling in Online Social Network (OSN) requires the frontal photographs of the users as thier Profile Pictures in Multi-Social Networking. The existing algorithms are ineffective in detecting the facial features like eyes, mouth and nose on the face appropriately, making it inefficient. This work proposes a novel approach to efficiently detect the facial features and improve the effectiveness of face detection and recognition by bifurcating the detected face horizontally, vertically and cropping it. The algorithm is effectively run only on the portion of the detected face Bounded Box (BB) and area to generate bounded boxes of other facial objects and later the Euclidian Distance (ED) between those BBs with respect to that of the face is computed to get Logarithm of Determinant of Euclidian Distance Matrix (LDEDM) in Relative-Distance (RD) method and stored in the database. The LDEDM so computed is unique for every user under consideration and is further utilized for identity matching recognizing from the database. The results show that the Equal Error Rate (EER) is considerably low indicating accurate threshold fixation for better performance with the proposed Relative Distance based User Profiling from Profile Picture (RDUP3) algorithm.

show abstract

A preliminary investigation on the sensitivity of COTS face recognition systems to forensic analyst-style face processing for occlusions

Cited by 19 publications

References 23 publications

Multi-class Fukunaga Koontz discriminant analysis for enhanced face recognition

Multi-class Fukunaga Koontz discriminant analysis for enhanced face recognition

RankGAN: A Maximum Margin Ranking GAN for Generating Faces

RDUP3: Relative Distance based User Profiling from Profile Picture in Multi-Social Networking

Contact Info

Product

Resources

About