Asian female face classification incorporating personal attractive preference

This paper proposes an Asian female face classification that incorporates personal attractive preference utilizing the reconstruction of PCA eigenface. Conventional PCA-based methods trained face images in all the classes as one huge training set. The produced eigenfaces contained general face information from all the classes that are not distinct to each class. To obtain eigenfaces with more specific face information for each class, the proposed method handles each class separately. Then, the similarity between the reconstructed image by utilizing eigenfaces of each class and the original image is measured and compared. From the experiments, while the accuracy results vary depending on the participants, the proposed method outperforms conventional PCA-based methods for all the participants, with a confidence level of 95% according to the Wilcoxon signed-rank test. In the 3-class classification, the proposed method achieves improvement in average accuracy ranging from 7.7% to 15.1% and in the 2-class classification, from 6.3% to 17.7%.