Novel local feature extraction for age invariant face recognition

Age variation is a major problem in the area of face recognition under uncontrolled environments such as pose variation, lighting effects, expression etc. Most of the works of this area have used discriminative feature descriptors. These discriminative feature descriptors are based on their fixed encoding which considers pixels of different radial widths for feature extraction and ignores some radii pixels which hold important discriminative information for age variation. Therefore, consideration of all the pixels of the local region is necessary for important feature extraction in the case of age invariant face recognition. This paper introduces a novel local feature descriptor to find difference pattern and dual directional relation pattern for age invariant face recognition. The proposed descriptor is applied over the preprocessed face images and its parts-periocular region i.e. left and right eye, mouth and nose region of a face image. The proposed difference pattern and dual directional relation pattern descriptors extract the texture features on the local region of a specified dimension. Chi-square metric has been used for finding the similarity between probe and gallery images. Evaluation of the proposed feature descriptor has been performed on two standard challenging datasets FGNET and MORPH for age invariant face recognition. The proposed descriptor performed well and outperformed to the existing age invariant face recognition state-of-the-art methods on FGNET dataset and also performed well on MORPH dataset.