Face and facial expression recognition using Extended Locality Preserving Projection

Face images of a person taken in varying expressions, orientations, lighting conditions are expected to be close to each other even under any mathematical transformation. These high dimensional face images are difficult to be recognized as faces of same person by machines in contrast to the humans. Many of the existing face recognition systems thus explicitly reduce the dimensions before performing recognition task. However, it is not guaranteed that varying faces of a single person could still be close in the lower dimensional space. Dimensionality reduction technique such as Extended Locality Preserving Projection (ELPP) not only reduces the dimension of the input data remarkably but also preserves the locality using neighbourhood information in the projected space. This paper deals with a face recognition system where ELPP is used to reduce the dimension of face images and hence uses ELPP coefficients as features to the classifier for recognition. In specific, two classifiers namely Naive Bayes classifier and Support Vector Machine are used. Results of face recognition of different data sets are highly impressive and at the same time results of facial expressions are encouraging. Experiments have also been carried out by taking a supervised version of ELPP (ESLPP).