Face Recognition

Burcu Kepenekci, F. Boray Tek and A. Alatan

The human identification systems based on the iris and fingerprints have already led to commercial products.Although they have very high identification rates these systems are not always appreciated by users as they require some close interaction with the machine often perceived as invasive. Face recognition may overcome these limitations.

Lawrence et al identify two areas in which face recognition can be used:

• Identification: Searching within a large database of faces to find a particular person (e.g. a police database). There are usually very few images per person in the database.
• Access Control (Authentication): Users not in the database cannot have access, which makes an area secure, as the condition for entry is no longer a pin number or a security card, but the individual's features. Processing has to be done in real-time. Multiple images per person are available for training.

In this study the goal is to identify particular people in real-time by using 2D gabor wavelet representations of facial images for face recognition.

Method is tested with three different face databases; Stirling, Purdue and ORL face databases. University of Stirling face database contains greyscale facial images of 35 people (18 female, 17 male), in 3 poses and 3 expressions with constant illumination conditions. In our early experiments we used three frontal images with different expressions for each of 35 persons, neutral views are placed in the gallery, images of smiling and speaking are used for test. In this database we achieve 100% correct recognition. In figure below two examples of recognized faces are shown.