An experimental framework for evaluation of facial feature extraction methods

Facial feature extraction is one of the hottest research topics in pattern recognition. Scholars have proposed numerous facial feature extraction methods based on various discriminant criteria, models, and algorithms. Each method has its own advantages and shortcomings. Unfortunately, till now there is no sound theoretical framework to evaluate their total performance. People have to resort to their experimental results. Since recognition accuracies and computational times of a particular facial feature extraction method in a certain simulation experiment are heavily depend on many factors such as, face image database, number of training samples per class, type of cross-validation, classifier, and parameter of the classifier used in the experiment. Thus, experimental design pays a key role in evaluation of their performance. In this paper we propose an experimental framework which can be used as a platform for a relatively fair comparison among facial feature extraction methods.