Interval type-2 fuzzy set induced fuzzy rank-level fusion for face recognition

Type-2 fuzzy set is extensively studied in the past due to its superiority over type-1 fuzzy set, especially while dealing with data with higher uncertainty and higher association amongst them. This paper proposes a framework by introducing interval type-2 fuzzy set induced fuzzy rank-level fusion for face recognition utilizing multi-feature vectors. It utilizes the outputs of a classifier as confidence factors. We address the wide intra-class variability issue by introducing interval type-2 fuzzy sets by using these confidence factors to generate secondary membership values from the intra-class face images, which are then reduced to a primary membership value. It also mitigates the influence of inter-class similarity by excluding the inter-class face images in the computation. Furthermore, multi-feature vectors for a face image are used to utilize the underlying different discriminant features, which also in turn address the issues of inter-class similarity. For each feature vector, we generate interval type-2 fuzzy set based fuzzy ranks corresponding to all classes. To reduce the complexity, top k fuzzy ranks are fused with corresponding membership values to obtain the fuzzy ranks. Likewise, the interval type-2 fuzzy set based multiple fuzzy ranks are obtained using multiple feature vectors. Finally, these fuzzy ranks are fused with the corresponding complemented confidence factors to get the final fuzzy ranks, based on which a face image is classified and recognized. The method is evaluated on several face databases and found to be superior to the many state-of-the-art methods.