Personalized feature combination for face recognition
Yuchun Fang, Yunhong Wang, Tieniu Tan · 2004
In this paper a novel personalized feature combination scheme is proposed for face recognition. ANFIS (adaptive neuro-fuzzy inference system) is adopted to form specialized feature representation for each subject by global and local features. For global features, we make a comparison between the two traditional global feature extraction schemes: PCA and LDA. The local features are extracted with wavelet packet decomposition around the areas of facial features. Instead of the common way for different subjects, we realize a new representation that adapts to each individual. Such adaptability in feature selection is inspired by the face recognition mechanism of the human visual system and results in an improved recognition rate.