• 1
  • 2
  • 3

 

 

Face recognition is very attractive because it includes a good compromise between people acceptance and reliability. Then, at BIPLab we developed three different approaches, two of them (FAST and FARO) exploiting the IFS theory, largely studied in still image compression and indexing, but not enough for the face recognition task, while the third one (FACE) is based on a local correlation index.

 

At last, the FAST algorithm is competitive with respect to a large set of other recognition methodologies in literature. In order to assess the performance of the FAST strategy, we compared it with several face recognition techniques from the literature. Tests have been performed using different protocols and measures, on several facial database sets available for evaluating face recognition algorithms. In particular, the databases available from FERET, Yale and MIT have been considered. These are freely available for non-commercial use and very popular in the face recognition community. From test results, it can be noted that FAST is more robust with respect to face expression changes and presence\absence of glasses, rather than to modification of illumination conditions. Moreover, FAST performs well over nearly all subjects in the database. This means that the strategy retains its high recognition rate independently of the subject’s characteristics.

Even if face authentication can be attempted with a high probability of success with consentient people, sometimes, it is necessary to deal with occlusions when the subject is wearing sunglasses, scarves and such. Then, an algorithm has been developed by the BIPLab for the recognition task, in order to cope with synthetic and natural occlusions. This method is based on IFS (Iterated Function Systems) theory too and is called FARO (Face Recognition in Occluded Conditions). One advantage is that the information used for the indexing and recognition task can be made local, and this makes the method more robust to possible occlusions. The distribution of similarities in the face image is exploited as a signature for the identity of the subject. The amount of information provided by each component of the face image has been assessed, first independently and then jointly. FARO addresses essentially the problem of face recognition when partial occlusions occur. The information about self-similarities in the face image has been located in four different regions in order to guarantee the independence among face objects such as eyes, nose and mouth. Also changes in expression can be seen as facial occlusions, which however affect the authentication process less heavily than accessories such as scarves or sunglasses.

 

Experimental tests have been conducted in order to assess the performance of FARO with respect to changes in expression and presence/absence of synthetic and natural occlusions. FARO has been tested with several sets of face images extracted from two largely used database: FERET and AR Faces. Because of the irregular nature of the images in these databases, a normalization procedure is required. In the case of FARO, face images are pre-processed in order to assure that eyes and mouth are always located in the same prefixed position, while no warping process has been applied. Experimental results shown the robustness of FARO with respect to synthetic occlusions (black squares randomly localized on the image) and natural occlusions such as scarves and sunglasses.