Bezier Cohort Fusion in Doubling States for Human Identity Recognition with Multifaceted Constrained Faces

Cohort selection benefits a biometric system by providing the information collected from non-match templates, whereas fusion benefits a system by combining information collected from different sources or from same source in different ways. The benefits of both approaches are availed here by proposing a cohort selection technique which is exploited prior to fusion and after fusion of matching scores for a face recognition system. Two robust facial features, viz. scale-invariant feature transform and speeded up robust features, are used here. This study presents a novel way of fusion based on cohort selection unlike the traditional levels of fusion (i.e., sensor, feature, match score, rank and decision level fusions). Cohort-based fusion is performed in two different fashions—pre-cohort fusion and post-cohort fusion. In case of early fusion, fusion rules like sum, max, min and average rules are applied before cohort selection to be performed. In contrast, the cohort selection is followed by the fusion in post (or late)-cohort fusion. Union operation is applied as late fusion rule. The matching scores are normalized by T-norm cohort score normalization technique prior to be compared with the threshold value to govern the decision of acceptance by the system. The experiments are carried out on FEI and the Look-alike (IIIT Delhi) face databases. The outcomes of the proposed method are looked to be encouraging and much convincing over non-cohort systems and state-of-the-art methods.