Analysis Of Effect Of Optimization Algorithm On Pose And Occlusion Invariant Feature For Face Recognition
Feature selection and representation play a significant role in the design of machine learning systems. Having a
poor feature representation can severely limit the performance of learning algorithms. Also machine learning applications
require features which are invariant to scale, rotation, illumination, expression and occlusion. This paper describes Fuzzy_
Bat clustering based Feature selection and optimization of an extensive feature set obtained by combining local scale and
rotation invariant feature, appearance based feature and entropy based feature for recognizing faces from video with focus on
pose and occlusion invariance. The fuzzy_ Bat clustering algorithm is used to search the optimal features from the feature
space based on a well-defined criterion. The effect of the classifier and optimization algorithm on recognition accuracy is
studied with Honda UCSD video database dataset1. Experimental results show that the Fuzzy_ Bat clustering based feature
selection and representation perform well with optimized features when compared to fuzzy rule based and Fuzzy C_ means
based algorithms. Also it has been observed from simulation results that using a squared Euclidean distance based classifier
enhances recognition rate when compared to Euclidean distance based classifier.
Keywords— Entropy, Feature selection, Feature Optimization, Fuzzy_ Bat clustering, Squared Euclidean Distance.