Comparing Human and Automatic Face Recognition Performance
Bibliographic record
Abstract
Face recognition technologies have seen dramatic improvements in performance over the past decade, and such systems are now widely used for security and commercial applications. Since recognizing faces is a task that humans are understood to be very good at, it is common to want to compare automatic face recognition (AFR) and human face recognition (HFR) in terms of biometric performance. This paper addresses this question by: 1) conducting verification tests on volunteers (HFR) and commercial AFR systems and 2) developing statistical methods to support comparison of the performance of different biometric systems. HFR was tested by presenting face-image pairs and asking subjects to classify them on a scale of "Same," "Probably Same," "Not sure," "Probably Different," and "Different"; the same image pairs were presented to AFR systems, and the biometric match score was measured. To evaluate these results, two new statistical evaluation techniques are developed. The first is a new way to normalize match-score distributions, where a normalized match score t is calculated as a function of the angle from a representation of [false match rate, false nonmatch rate] values in polar coordinates from some center. Using this normalization, we develop a second methodology to calculate an average detection error tradeoff (DET) curve and show that this method is equivalent to direct averaging of DET data along each angle from the center. This procedure is then applied to compare the performance of the best AFR algorithms available to us in the years 1999, 2001, 2003, 2005, and 2006, in comparison to human scores. Results show that algorithms have dramatically improved in performance over that time. In comparison to the performance of the best AFR system of 2006, 29.2% of human subjects performed better, while 37.5% performed worse.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.008 | 0.020 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.000 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".