Facial Face Recognition Based On High-Performance Topographical Features Selected Using Adaptive Feature Selection Method

Abstract

In this paper, a new features selection method is represented to enhance a feature type that is already
adopted in previous classification researches. By studying successive differences between image values of
face image, a relatively huge set of topographical features (TGF) are extracted from assigning each value
image pixel to the related feature. An Efficient Feature Selection for face recognition (EFS) method is
proposed to study and analyze TGF features. Such method used standard measures used to determine
feature performance adopted in standard methods, in addition to the major concept of Principal Component
Analysis (PCA) in dimension reduction for determining efficient features. EFS analyzed the dispersion of
yielded values for each feature within images of the dataset in order to determine the performance level of
each feature. Eventually, the level of determined performance is assigned to the related features which are
sorted in descending manner. Based on a controlled threshold, the highest set of performance values, which
are the best expected set of features, are selected as candidate features for Support Vector Machine (SVM)
classifier. Collecting proposed features and the selection method yielded encouraging results up to
(94.12%) as classification accuracy with considerable differences comparing with the State of Art.