Modified feature extraction techniques to enhance face and expression recognition

Many researchers consider the face’s orientation and illumination while considering the face recognition process to obtain a reasonable recognition rate. They also consider extracting expression-targeted features for expression recognition. Understanding feelings via face and expression recognition would monitor and control office environments effectively to help manage people. This research aims to identify specific features, including face orientation, to improve the recognition rate. The state-of-the-art solution consists of face landmark, affine transformation, and modified feature extraction techniques using the Attention module and Local Binary Pattern (LBP) that address the face orientation, illumination, and targeted features issues. In this paper, a modified solution introduces an improved face recognition system using a pre-processing layer like a face landmark and affine transformation, a modified attention module, and circular derivative LBP in the state-of-the-art solution. The new proposed model also adds two different ways, shear and zoom, to the existing base method for data augmentation. Four datasets are considered in evaluating the proposed system: MMA, JAFFE, FER2013, and CKplus. The proposed system achieves an average accuracy of 95.63% compared to the state-of-the-art solution's accuracy (93.72%). The proposed system also reduces the time from 32.54ms to 26.45ms. Besides, the proposed system produces a relatively viable average F1 score and AUC of 87.66 and 88.98, respectively, higher than the state-of-the-art solution. The results prove that the proposed system increases the face recognition rate by applying a modified feature extraction process as a crucial component and eliminating inter-and intra-class separation issues.