A Novel Facial Expression Recognition System using BMCSA Based Adaptive Neuro-Fuzzy Inference System

In Emotion Recognition (ER), automatic Facial Expression Recognitions (FER) is actually a dynamically emerging research. Recently, the focus is given to two main issues (i.e.) overfitting because of the devoid of adequate training data as well as expression-unassociated variations, for instance, head pose, illumination, along with identity bias. Here, Brownian motion based Crow Search Algorithm for Adaptive Neuro Fuzzy Inference System (BMCSA-ANFIS) is proposed for doing the FER of human images. It mostly concentrates on recognizing an individual's Facial Expressions (FE) as of a single image. The proposed method comprises '4' stages. Initially, the inputted facial images are taken from a database that is openly available. Then, the images (inputted) are pre-processed by performing filtering (noise removal) along with energy normalization. After that, the feature extraction is performed utilizing Local directional patterns (LDP) along with Appearances Model (AAM) centered Scale Invariants Features Transform (SIFT) that extracts the features as of the preprocessed images for doing FER. Then, the system utilizes the Information Gain based Linear Discriminants Analysis (IG-LDA) to lessen the extracted features (dimensionality reduction). Subsequent to Feature Reduction (FR), the reduced features are rendered to BMCSA-ANFIS for classification that classifies the result of the inputted images into '6' fundamental emotions: happy, angry, disgust, surprise, sad, together with neutral by considering the reduced features. The experimentation's outcomes are evaluated and that clearly exhibited that the proposed techniques perform superiorly to other existent algorithms for FER.