Human Face Recognition
|Challenges for face recognition|
Challenges for face recognition
The face is a three-dimensional (3D) object. Its appearance is determined by the shape as
well as texture of the face. Broadly speaking, the obstacles that a face recognition system must overcome are differences in appearance due to variations in illumination, viewing angle, facial expressions, occlusion and changes over time. Using 2D images for face recognition, the intensities or colours of pixels represent all the information that is available and therefore, any algorithm needs to cope with variation due to illumination explicitly. The human brain seems also to be affected by illumination in performing face recognition tasks (Hill et al., 1997). This is underlined by the difficulty of identifying familiar faces when lit from above (Johnston et al., 1992) or from different directions (Hill and Bruce, 1996). Similarly it has been shown that faces shown in photographic negatives had a detrimental effect on the identification of familiar faces (Bruce and Langton, 1994). Further studies have shown that the effect of lighting direction can be a
determinant of the photographic negative effect (Liu et al., 1999). As a result, positive faces, which normally appear to be top-lit, may be difficult to recognize in negative partly because of the accompanying change in apparent lighting direction to bottom-lit. One explanation for these findings is that dramatic illumination or pigmentation changes interfere with the shape-from-shading processes involved in constructing representations of faces. If the brain reconstructs 3D shape from 2D images, it remains a question why face recognition by
humans remains view point dependent to the extent it is. One of the key challenges for face recognition is the fact that the difference between two images of the same subject photographed from different angles is greater than the differences between two images of different subjects photographed from the same angle. It has been reported that recognition rates for unfamiliar faces drop significantly when there are different viewpoints for the training and test set. More recently, however, there has been debate about whether object recognition is viewpoint-dependent or not. It seems that the brain is good at generalizing from one viewpoint to another as long as the change in angle is not extreme. For example, matching a profile viewpoint to a frontal image is difficult, although the matching of a three-quarter view to a frontal seems to be less difficult. There have been suggestions that the brain might be storing a view-specific prototype abstraction of a face in order to deal with varying views. Interpolation-based models , for example, support the idea that the brain identifies faces across different views by interpolating to the closest previously seen view of the face. Another key challenge for face recognition is the effect of facial expressions on the appearance of the face. The face is a dynamic structure that changes its shape non-rigidly since muscles deform soft tissue and move bones. Neurophysiologic studies have suggested that facial expression recognition happens in parallel to face identification (Bruce, 1988). Some case studies in prosopagnostic patients show that they are able to recognize expressions even though identifying the actor remains a near-impossible task. Similarly, patients who suffer from organic brain syndrome perform very poorly in analyzing expressions but have no problems in performing face recognition. However, the appearance of the face also changes due to aging and people’s different lifestyles. For example, skin becomes less elastic and more loose with age, the lip and hair-line often recedes, the skin color changes, people gain or lose weight, grow a beard, change hairstyle etc. This can lead to dramatic changes in the appearance of faces in images. A final challenge for face recognition is related to the problem of occlusions. Such occlusions can happen for a number of reasons, e.g. part of the face maybe occluded and not visible when images are taken from certain angles or because the subject grew a beard, is wearing glasses or a hat.
Want To Know more with
Contact for more learning: webmaster@freehost7com
The contents of this webpage are copyrighted © 2008 www.freehost7.com
All Rights Reserved.