An improved sparse code representation using local matching for deterministic face authentication

The new framework proposed in this paper provides an insight into the problem of face authentication (verification) in unconstrained environment. This unconventional method extracts and represents the microstructures and local features of a given face image by greedy approach and sparse code respectively. This gives a stable and discriminative local descriptor for each patch that hinge on the local patches and learned dictionary. Dictionary is learned from the local patches of each facial patch (component) selected using greedy approach and optimality check. Compared to the previous sparse representation based methods, new method is actually a fusion of local component and region approach. The proposed method outperforms the existing method and gives an accuracy of 99% which is demonstrated through extensive experiments conducted on publically available and challenging LFW dataset.