摘要：Face image variations such as expression and pose radically increase the intra-class variations which affect the performance of feature extraction methods. It is desirable to extract more robust local discriminative features to effectively represent such face variations. This paper proposes a novel facial feature extraction method which utilizes interpolation-based directional wavelet transform (DIWT) and local binary patterns (LBP). An efficient direction assessment method based on quadtree partitioning is implemented to facilitate adaptive direction selection in the local regions from the face images to obtain DIWT sub-bands. LBP histogram features are extracted from selected top-level DIWT sub-bands to obtain local descriptive feature set. Experimental results are simulated on ORL database, GT database, and FEI database. A comparison with various contemporary methods which include holistic, local descriptors and LBP-based non-adaptive multiresolution analysis methods illustrate the efficacy of the proposed method.
