Face recognition using scale-adaptive directional and textural features

• We propose a face descriptor based on a combination of directional and textural features.
• Discriminative and nearly illumination invariant directional features are introduced.
• Pyramid partitioning is used to capture local as well as holistic features.
• Experiments performed on six standard face datasets shows robustness of proposed descriptor.
• Algorithm achieves nearly perfect recognition rate on a number of standard face datasets.

A novel approach to face recognition problem using directional and texture information from face images, is proposed in this paper. In order to capture the directionality, specially designed using local polynomial approximation technique, scale adaptive digital filters are used. For texture features extraction, a low dimensional and computationally effective local descriptor is utilized. Textural and directional features are captured at the holistic and part based levels resulting in a robust face descriptor. The proposed method is tested on a number of standard test face datasets (ORL, XM2VTS, Extended Yale, CMU-PIE, AR, and FERET) for different scenarios and its performance is compared with several state-of-the-art techniques.