Face recognition has become one of the most addressed pattern recognition problems due to its importance as a natural biometric trait and due to its role in human computer interfaces.


Face detection has wide range of applications such as automatic face recognition, human­ machine interactions, surveillance, etc. In recent years there has been substantial progress on detection schemes based on appearances of faces. Holistic approaches have been dominating the face recognition research since the beginning of 1990s. On the other hand , recently, local appearance based face recognition approaches have attracted a growing interest. In salient local regions, such as the eye regions, are used to perform modular eigenfaces based face recognition.


In  recent  face  recognition  system, the  face  or  image  is  considered  as  block  of rectangular images. HaZlln Kemal Ekene1(2009),  in his paper on "Block  Selection in the Local Appearance-based Face Recognition Scheme, describes that the face image is divided . into  rectangular  smaller  sub-images  without  considering  any  specific  regions,  and  the eigenfaces approach is then performed on each of these sub-images. The local facial regions are  located  by  a  Support  Vector  Machine  (SVM)  and  the  combined  local  features  are classified again with SVM. The face image is partitioned into several local regions and each local region is represented by Linear Discriminant Analysis (LOA). To combine the features extracted from each local region, another LOA is used.   Along with these two techniques, Discrete Cosine Transform(DCT) is also used. OCT plays the main role for face detection in this dissertation. DCT is combined with a Block Weighting Scheme. The Block Weighting Scheme measures the contribution of each block in a face image.


The basic idea is: clean blocks should impact the result more than the occluded one. For each block in one image, the probability of the block belonging to each class, rather than the classification result is computed. These probabilities are combined to obtain classification result for the whole image.