by Jennifer R
The authors talk about the growing interest in facial recognition databases and how they feel the databases currently being studied by researchers do not mimic real world scenarios. Researchers should take into account that lighting may not be ideal for a clear image, or that the subject may be moving or looking in a direction that may obscure their faces or result in a poor-quality image. The authors also point out the distance of the subject from the camera will affect the image, as well as the type of camera being used. They created their own database, by setting up six cameras of differing quality, with one high-one quality camera being set aside for mug shots. The article details the criteria they used for naming images, such that the names were unique and carried information about the image. Participants were to follow a set of instructions ensuring that they walked past the cameras in the same way as everyone else. They also conducted image capturing at night. The purpose of such a database is to explore the problems of facial recognition programs that have yet to be addressed and to point out the factors of the real-world environment can greatly affect recognition performance.
This article is interesting in that it points out there are flaws in using databases that query in the form of image comparison. Images stored in databases generally are in ideal condition. They are clear, well-defined and of good quality. The article shows the differences between the captured image and database can be influenced by environmental factors. What does this mean for the quality of the comparison? Will this affect the accuracy of the results? I can think of another database that could suffer similar problems: fingerprint databases.
Source: Ggic, M., Delac, K., Grgic, S. (2009) SCFace – Surveillance Cameras Face Database. Multimedia Tools and Applications, 51. Retrieved May 15, 2012 from http://www.scface.org/SCface%20-%20Surveillance%20Cameras%20Face%20Database.pdf