ABSTRACT
Face recognition has become a widely researched field due to its extensive applications, especially in the fields of security and surveillance. In recent years, various techniques have been propounded, one of which is the sparse representation-based classification (SRC). It is based on a sparse representation calculated by l1-minimization and provides classification for recognition. It assumes that the coding residual follows Gaussian or Laplacian distribution, which is, generally, not the case for practical systems. Thus, recognition is inaccurate when there is illumination variation, expression change, or occlusion involved. The proposed method is a face coding model called regularized robust coding (RRC), which can robustly regress a signal using regularized regression coefficients. An iteratively reweighted regularized robust coding (IR3C) algorithm is employed to handle the RRC model efficiently. The weight function is developed via MAP (Maximum A Posteriori) estimation. Specifically, RRC with l1-norm regularization can accomplish high recognition rates, as it yields a sparse solution and automatically performs feature selection. Also, it is robust to outliers, as it assigns insignificant features with zero weight and useful features with nonzero weight. Experiments performed on the Extended Yale B database indicate that superior performance is obtained with RRC-l1 compared to other sparse representation-based approaches in treating face occlusion, corruption, lighting, and expression variations. The important results are the 99.8% recognition rate for faces without occlusion and 96.7% for facial images with 40% block occlusion. Thus, the advantage of the proposed system is its ability to recognize faces even in the presence of poor lighting conditions, expression variations, and occlusion (sunglasses, scarves, etc.). Facial biometrics can be integrated with physical devices, like cameras, alarms, and other suitable peripherals, to identify people entering a building and issue an alarm if an unauthorized person attempts to enter. Face recognition-based security is especially useful for organizations and in smart cities that need to maintain tight control on those entering their facilities.
