Nowadays, applications in the field of surveillance, banking and multimedia equipment are
becoming more important, but since each application related to face analysis demands
different requirements on the analysis process, almost all algorithms and approaches for face
analysis are application dependent and a standardization or generalization is quite difficult.
For that reason and since many key problems are still not completely solved, the face analysis
research community is still trying to cope with face detection and recognition challenges.
Although emulating human vision system would be the ideal solution, it is a heuristic and
complicated approach which takes into account multiple clues such as textures, color, motion
and even audio information. Therefore, and due to the fast evolution of technology that makes
it possible, the recent trend is moving towards multimodal analysis combining multiple
approaches to converge to more accurate and satisfactory results.
Contributions to specific face detection and recognition applications are helpful to enable the
face analysis research community to continue building more robust systems by concatenating
different approaches and combining them. Therefore, the aim of this research is to contribute
by exploring the Local Binary Patterns operator, motivated by the following reasons. On one
hand, it can be applied to face detection and recognition and on the other hand due to its
robustness to pose and illumination changes.
Local Binary Patterns were first used in order to describe ordinary textures and, since a face
can be seen as a composition of micro textures depending on the local situation, it is also
useful for face description. The LBP descriptor consists of a global texture and a local texture
representation calculated by dividing the image into blocks and computing the texture
histogram for each one. The global is used for discriminating the most non-face objects
(blocks), whereas the second provides specific and detailed face information which can be
used not only to select faces, but also to provide face information for recognition.
The results will be concatenated in a general descriptor vector, that will be later used to feed
an adequate classifier or discriminative scheme to decide the face likeness of the input image
or the identity of the input face in case of face recognition. It is in that stage where this
research will focus, first evaluating more simple classification methods and then trying to
improve face detection and recognition ratios by trying to eliminate features vector
redundancy.
