PSI (ψ) Invariant Features for Face Recognition

Over last few decades, mathematics has played a crucial role in developing efficient algorithms for Face Recognition (FR) used in biometric systems. FR using Machine Learning (ML) techniques has impacted FR systems tremendously, towards efficient and accurate models for FR. Existing FR systems used in biometrics use ML techniques to learn patterns in the images by extracting various features from them and often require pre-processed face image data for the learning process. In this paper, we have used various pre-processing techniques and compared them in the deployed FR framework. It was observed that the Steerable Pyramid (SP) filter was the most efficient pre-processing technique among all techniques used for pre-processing in this work. Though existing feature extraction methods such as SIFT (Scale-Invariant Feature Transform), SURF (Speeded-Up Robust Features), ORB (Oriented FAST and Rotated BRIEF) have been used in the past, they have not been accurate enough in various vision based biometric systems. Hence, a novel PSI (Pose Scale and Illumination) invariant SURF-RootSIFT technique is proposed by extending the well known SIFT-RootSIFT feature extraction technique which is achieved by calculating the Bhattacharya Coefficient between the feature vectors. A framework which uses the proposed novel feature extraction technique is deployed in this work. This paper demonstrates that the novel SURF-RootSIFT based framework is proven to perform more accurately and efficiently than the other techniques, with 99.65, 99.74 and 97.93% accuracy on the Grimace, Faces95 and Faces96 databases respectively.