An Attempt to Classify Asian Female Faces Based on Personal Attractive Preference

Abstract

This paper attempts to classify Asian female faces with personal attractive preference utilizing PCA eigenface reconstruction ability. Conventional PCA-based methods combine face images in both attractive and not attractive, group into one huge training set to construct the eigenfaces. On the contrary, the proposed method constructs eigenfaces separately for both attractive and not attractive group. Hence, the eigenfaces will have more specific face features for each group respectively. Then, the similarity between the reconstructed image by eigenfaces of each group with the original image is measured and compared. The method yields average accuracy rate of 84.1% using Euclidean distance as the similarity measurement. It achieves improvement of accuracy by 12.7% and 7.7% if compared to Turkmen method with KNN classifier and Eisenthal pixel image SVM classifier, respectively.