Enhanced face recognition using rank-level fusion with multi-feature vectors and interval type-3 fuzzy set

In the near past, interval type-3 fuzzy set theory has been introduced as a more effective means of addressing uncertainty in decision-making than its fuzzy set counterparts. This paper suggests an approach to face recognition which integrates multiple feature vectors and interval type-3 fuzzy sets for fuzzy rank-level fusion. Use of multiple feature vectors resolves inter-class resemblances while incorporating interval type-3 fuzzy membership function in generating fuzzy rank reduces the uncertainty due to intra-class divergences. The output of the classifier is regarded as confidence factor. For each feature vector, the fuzzy ranks are generated by formulating an interval type-3 fuzzy membership function for each of top-ordered H classes based on confidence factors utilizing all training feature vectors and those of the training samples of the respective class. Multiple fuzzy ranks, built upon interval type-3 fuzzy set, are generated by employing of multiple feature vectors. To determine the final score of a class, summation of the fuzzy rank complements is combined with the complement of arithmetic mean of respective classifier’s output. Final scores are evaluated to conclude test image's class. Two publicly accessible face databases are used to evaluate the method and it has shown that the solution, advocated here, exhibits an improved effectiveness over its existing related counterparts.