Face recognition in low-resolution images under small sample conditions with face-part detection and alignment

In order to support human users with daily tasks, computers need tobe aware of human presence and identity. Today, it is very easy toobtain visual information using, e.g. smart phones. Therefore, AIbased algorithms for face detection and recognition are becomingincreasingly relevant. Face recognition is a very easy and convenienttechnique compared to other biometrics methods like finger print oriris recognition. It does, however, still not perform very robustwhen compared to humans and is less reliable than alternativebiometrics methods.

This is due to many variations existing in photographs and videos (e.g. illumination, pose), which create challenging problems for a facerecognition algorithm. We focused mainly on three challenges in thisthesis. The first two are localization and rotational alignment of thefaces, which are preprocessing steps before the recognition. The thirdstep is face identification itself, based on very little training data.For localization, we developed an eye detector localizing eye centersafter face detection. Rotational alignment is done using the anglesfrom these eye centers. For the single and small sample problems, weproposed two novel methods to handle the insufficient amount of data.The experiments lead to two important conclusions: First, a big genericdataset can help to improve recognition performance significantly forthe new target faces. Second, if the number of photographs is limited,then smart sampling of many patches helps improving identificationaccuracy.

Summarizing, although our research contributes to solving faceidentification in small sample conditions, further research isnecessary to obtain more robust results.

Dissertation: http://hdl.handle.net/11370/40793c4e-6c20-4fa9-bd85-7a99fcb6a5c0