Characterizing the Variability in Face Recognition Accuracy Relative to Race

Krishnapriya, K S; Vangara, Kushal; King, Michael C.; Albiero, Vítor; Bowyer, Kevin W.

doi:10.48550/arxiv.1904.07325

Cited by 2 publications

(3 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the second dataset, there are four photographs of twelve different Black female celebrities. Since factors such as the quality of face images and the ratio of light affect the results [50], datasets were compiled from images with similar quality and light intensity.…”

Section: Deepface Framework and Facial Recognition Models Deepface Fr...mentioning

confidence: 99%

“…We know that facial recognition models need to be trained with balanced data sets in terms of gender and race so as not to cause bias. At the same time, factors such as the quality of facial images, and shadow and light ratio directly affect the results produced by face recognition models [50]. Although we use trained CNN models, these standards are also important for determining appropriate threshold values.…”

Section: The Line-up Applicationmentioning

confidence: 99%

See 1 more Smart Citation

A method for analyzing suspect-filler similarity using convolutional neural networks

AYDIN>¹,

2022

Communications Faculty of Sciences University of Ankara Series A2-A3 Physical Sciences and Engineering

View full text Add to dashboard Cite

Eyewitness misidentifications are one of the leading factors in wrongful convictions. This study focuses on the structure of the lineups, which is one of the factors that cause misidentification, and the use of artificial intelligence (AI) technologies in the selection of fillers to be included in the lineups. In the study, AI-based face recognition systems are used to determine the level of similarity of fillers to the suspect. Using two different face recognition models with a Convolutional Neural Network (CNN) structure, similarity threshold values close to human performance were calculated (VGG Face and Cosine similarity = 0.383, FaceNet and Euclidean l2 = 1.16). In the second part of the study, the problems that are likely to be caused by facial recognition systems used in the selection of fillers are examined. The results of the study reveal that models responsible for facial recognition may not suffice alone in the selection of fillers and, an advanced structure using CNN models trained to recognize other attributes (race, gender, age, etc.) associated with similarity along with face recognition models would produce more accurate results. In the last part of the study, a Line-up application that can analyze attributes such as facial similarity, race, gender, age, and facial expression, is introduced.

show abstract

Section: Deepface Framework and Facial Recognition Models Deepface Fr...mentioning

confidence: 99%

Section: The Line-up Applicationmentioning

confidence: 99%

A method for analyzing suspect-filler similarity using convolutional neural networks

AYDIN>¹,

2022

Communications Faculty of Sciences University of Ankara Series A2-A3 Physical Sciences and Engineering

View full text Add to dashboard Cite

show abstract

“…Other datasets, such as PPB [7] and Diversity in Faces [32], use phenotype annotations in lieu of racial categories. Audits of algorithms, APIs, and software [39], [17], [27] have found statistically significant disparities in performance across groups and have been critical to bringing racial disparities to light.…”

Section: Algorithmic Fairness and Datasetsmentioning

confidence: 99%

One Label, One Billion Faces

Khan

2021

Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency

View full text Add to dashboard Cite

Computer vision is widely deployed, has highly visible, societyaltering applications, and documented problems with bias and representation. Datasets are critical for benchmarking progress in fair computer vision, and often employ broad racial categories as population groups for measuring group fairness. Similarly, diversity is often measured in computer vision datasets by ascribing and counting categorical race labels. However, racial categories are ill-defined, unstable temporally and geographically, and have a problematic history of scientific use. Although the racial categories used across datasets are superficially similar, the complexity of human race perception suggests the racial system encoded by one dataset may be substantially inconsistent with another. Using the insight that a classifier can learn the racial system encoded by a dataset, we conduct an empirical study of computer vision datasets supplying categorical race labels for face images to determine the cross-dataset consistency and generalization of racial categories. We find that each dataset encodes a substantially unique racial system, despite nominally equivalent racial categories, and some racial categories are systemically less consistent than others across datasets. We find evidence that racial categories encode stereotypes, and exclude ethnic groups from categories on the basis of nonconformity to stereotypes. Representing a billion humans under one racial category may obscure disparities and create new ones by encoding stereotypes of racial systems. The difficulty of adequately converting the abstract concept of race into a tool for measuring fairness underscores the need for a method more flexible and culturally aware than racial categories. CCS CONCEPTS • Social and professional topics → Race and ethnicity; • Computing methodologies → Computer vision.

show abstract

Characterizing the Variability in Face Recognition Accuracy Relative to Race

Cited by 2 publications

References 0 publications

A method for analyzing suspect-filler similarity using convolutional neural networks

A method for analyzing suspect-filler similarity using convolutional neural networks

One Label, One Billion Faces

Contact Info

Product

Resources

About