Face Recognition Algorithms Surpass Humans Matching Faces Over Changes in Illumination

O'Toole, Alice J.; Phillips, P. Jonathon; Jiang, Fang; Ayyad, Julianne; Pénard, Nils; Abdi, Hervé

doi:10.1109/tpami.2007.1107

Cited by 178 publications

(109 citation statements)

References 23 publications

Supporting

Mentioning

102

Contrasting

Unclassified

Order By: Relevance

“…However, the adequacy of this ''pigmentation hypothesis'' has been challenged by data showing that hue negation, which also results in unnatural pigmentation (making the entire face look bluish-green), has little effect on recognition performance (11). It is also unclear whether observers can usefully extract pigmentation information across different illumination conditions (17,18). We propose an alternative account of negation-induced impairment.…”

mentioning

confidence: 87%

Role of ordinal contrast relationships in face encoding

Gilad

Meng²,

Sinha³

2009

Proc. Natl. Acad. Sci. U.S.A.

116

View full text Add to dashboard Cite

What aspects of facial information do we use to recognize individuals? One way to address this fundamental question is to study image transformations that compromise facial recognizability. The goal would be to identify factors that underlie the recognition decrement and, by extension, are likely constituents of facial encoding. To this end, we focus here on the contrast negation transformation. Contrast negated faces are remarkably difficult to recognize for reasons that are currently unclear. The dominant proposals so far are based either on negative faces' seemingly unusual pigmentation, or incorrectly computed 3D shape. Both of these explanations have been challenged by recent results. Here, we propose an alternative account based on 2D ordinal relationships, which encode local contrast polarity between a few regions of the face. Using a novel set of facial stimuli that incorporate both positive and negative contrast, we demonstrate that ordinal relationships around the eyes are major determinants of facial recognizability. Our behavioral studies suggest that destruction of these relationships in negatives likely underlies the observed recognition impairments, and our neuro-imaging data show that these relationships strongly modulate brain responses to facial images. Besides offering a potential explanation for why negative faces are hard to recognize, these results have implications for the representational vocabulary the visual system uses to encode faces.contrast negation ͉ face perception ͉ fMRI ͉ neural representation ͉ object recognition I n principle, a contrast negated image is exactly as informative as its positive counterpart; negation perfectly preserves an image's 2D geometric and spectral structure. However, as anyone who has had to search through a roll of negatives for a snapshot of a particular person knows, this simple operation has dramatically adverse consequences on our ability to identify faces, as illustrated in Fig. 1 (1-6). Exploring the causes of this phenomenon is important for understanding the broader issue of the nature of information the visual system uses for face identification.Several researchers have hypothesized that negated face-images are hard to recognize because of the unnatural shading cues in negatives, which compromise shape from shading processes (7-11). The resulting problems in recovering veridical 3D facial shape are believed to impair recognition performance. Although plausible, it is unclear whether this explanation is a sufficient one, especially in light of experimental results showing preserved recognition performance in the absence of shading gradients (12), and theories of face recognition that are based on the use of 2D intensity patterns rather than recovered 3D shapes (13,14). Another prominent hypothesis is that negation causes faces to have unusual pigmentation (15, 16). However, the adequacy of this ''pigmentation hypothesis'' has been challenged by data showing that hue negation, which also results in unnatural pigmentation (making the entire face l...

show abstract

mentioning

confidence: 87%

Role of ordinal contrast relationships in face encoding

Gilad

Meng²,

Sinha³

2009

Proc. Natl. Acad. Sci. U.S.A.

116

View full text Add to dashboard Cite

show abstract

“…Even though it was recently demonstrated that machines have started to become more efficient than humans in face recognition in unconstrained conditions [207,208], humans still largely surpass machines in face detection [209]. Nevertheless, it is exciting to see face detection techniques be increasingly used in real-world applications and products.…”

Section: Discussion Future Work and Conclusionmentioning

confidence: 99%

A survey on face detection in the wild: Past, present and future

Zafeiriou

Cha

Zhang

2015

Computer Vision and Image Understanding

386

167

View full text Add to dashboard Cite

“…Former studies have shown that, when a large number of representative training images are available, computer algorithms are able to recognize even better than humans [6], [7], [25], [27], [33]. These algorithms, before recognizing, represent the face in the feature space, and perhaps because they are able to extract information from training images about the changes caused by different conditions they outperform humans in recognition [25].…”

Section: Introductionmentioning

confidence: 99%

“…These algorithms, before recognizing, represent the face in the feature space, and perhaps because they are able to extract information from training images about the changes caused by different conditions they outperform humans in recognition [25].…”

Section: Introductionmentioning

confidence: 99%

Changes in Facial Expression as Biometric: A Database and Benchmarks of Identification

Haamer

Kulkarni

Imanpour

et al. 2018

2018 13th IEEE International Conference on Automatic Face &Amp; Gesture Recognition (FG 2018)

View full text Add to dashboard Cite

Abstract-Facial dynamics can be considered as unique signatures for discrimination between people. These have started to become important topic since many devices have the possibility of unlocking using face recognition or verification. In this work, we evaluate the efficacy of the transition frames of video in emotion as compared to the peak emotion frames for identification. For experiments with transition frames we extract features from each frame of the video from a fine-tuned VGG-Face Convolutional Neural Network (CNN) and geometric features from facial landmark points. To model the temporal context of the transition frames we train a Long-Short Term Memory (LSTM) on the geometric and the CNN features. Furthermore, we employ two fusion strategies: first, an early fusion, in which the geometric and the CNN features are stacked and fed to the LSTM. Second, a late fusion, in which the prediction of the LSTMs, trained independently on the two features, are stacked and used with a Support Vector Machine (SVM). Experimental results show that the late fusion strategy gives the best results and the transition frames give better identification results as compared to the peak emotion frames.

show abstract

Face Recognition Algorithms Surpass Humans Matching Faces Over Changes in Illumination

Cited by 178 publications

References 23 publications

Role of ordinal contrast relationships in face encoding

Role of ordinal contrast relationships in face encoding

A survey on face detection in the wild: Past, present and future

Changes in Facial Expression as Biometric: A Database and Benchmarks of Identification

Contact Info

Product

Resources

About