Gabor Filter Based Face Recognition Using Non-frontal Face Images

Abstract

Face recognition has immense real world applications in the field of computer vision and a challenging task especially when the frontal face images are not available to train the classifiers. In this paper by regulating the scale and orientation parameters of Gabor Filters, we obtain high dimensional features from the face images with different poses. To classify the images, first we partition the images using k-means clustering algorithm where k varies from 6 to 8 for different databases representing pose variations of input images. Based on the clustering we assign class labels to the training data set for recognizing non-frontal face images with variant poses. To reduce the complexity of the system, different statistical properties of the features like variance, entropy, and correlation coefficient are analysed to select significant features only. Removal of irrelevant features, effectively reduces dimensionality of the feature space without sacrificing accuracy which is 94.47%. The proposed approach performs better compare to the existing methods, with and without feature selection algorithm.