FERET
Technical information:
| Full name | Database of registering prohibitions on entry |
|---|---|
| Country | |
| Contents | Facial Images |
| Images | 14,126 |
| Individuals | 1,199 |
| Runs database software | |
| URL"URL" is a type and predefined property provided by Semantic MediaWiki to represent URI/URL values. | |
| Related Technology |
Developers and Users:
| Developed by | NIST DoD |
|---|---|
| Owning institution | |
| Custodian institution |
Description[edit | ]
The Face Recognition Technology (FERET) dataset was thus created with $6.5 million<ul><li>No units of measurement were declared for this property.</li> <!--br--><li>"" is not declared as a valid unit of measurement for this property.</li></ul> of funding from the U.S. Department of Defense and the National Institute of Standards and Technology (NIST) to provide researchers the data they required to make progress in the field. In 15 photography sessions of the same set up between August 1993 and July 1996, images were collected in a semi-controlled environment (Phillips et al. 2000b)1. 2:2
The resulting benchmark began with 2,413 still face images, representing 856 individuals, and grew to contain 14,126 facial images of 1,199 individuals, available upon request. 2
At the moment of its release, it became the largest and most comprehensive effort to create a benchmark that would accurately compare and evaluate existing facial recognition algorithms (Phillips et al. 2000b). The large data effort coupled with a government sponsored effort to promote facial recognition algorithm development via competitions and research investments (Phillips et al. 2005) proved successful at igniting academic research interest in the field. 2
In 2000, given the success of the FERET database in stimulating research interest in facial recognition, commercial implementations of this technology began to appear and prompted the National Institute of Standards and Technology (NIST) to release the Facial Recognition Vendor Test (FVRT), a benchmark aimed at evaluating this emerging commercial systems. Even then, the expressed intended context of consideration for these tools were to be “applied to a wide range of civil, law enforcement and homeland security applications including verification of visa images, deduplication of passports, recognition across photojournalism images, and identification of child exploitation victims” (Ngan, Ngan, and Grother 2015).2
The creation of a larger, more substantial dataset allowed early computer vision methods, such as support vector machines (SVMs), simple convolutional neural networks (CNNs) and hidden Markov models (HMMs), to be applied to facial recognition with some promising results (Jafri and Arabnia 2009).
However, the commercialization attempts with these early methods revealed that even small environmental changes, such as in image illumination and a subject’s pose, could at this time be enough to obscure or distort the features required to make a match. Similarly, any unexpected change in their face - from aging to a new facial expression to partial occlusions, such as a scarf, mask, or pair of glasses – could cripple the performance of the technology (Sharif et al. 2017; Forczma´nski and Furman 2012; Yang, Kriegman, and Ahuja 2002).2
Given the dearth of available face data, certain strategies to better generalize across environments were still out of reach and it was considered that, at this time, “current algorithms for automatic feature extraction do not provide a high degree of accuracy and require considerable computational capacity” (Jafri and Arabnia 2009).2