Encode Locally, Discriminate Holistically: A Method Extracting Multi-resolution Features for Facial Age Estimation

To combine the global and local aging information of face images, this paper proposes a feature extraction method which can extract multi-resolution discriminative local features in a holistic way for age estimation. By using 2DPCA on the left side of the block matrixes, the face images are firstly transformed into lower dimensional representation matrixes whose rows can be seen as block-wise image representations in different resolutions. Discriminate features are then extracted holistically for each of the resolutions by viewing each row as a sample. Based on the weak estimator on each of the resolutions, an ensemble is finally constructed to perform the age estimation task. Experimental results confirm the reliability of the proposed method compared with related methods.