An Adaptive Ensemble Learning-Based Smart Face Recognition and Verification Framework with Improved Heuristic Approach

Abstract

Face recognition is an uncontrollable surveillance system but faces complex issues regarding minimal resolution, expression modification, and motion blur. Nowadays, researchers find more interest in face recognition analysis on pattern recognition and computer vision models. Several advancements in deep learning approaches for face recognition displayed a better efficacy rate in standard datasets. However, real-time face recognition approaches are applied and offer a poor efficacy rate. This modification leads to more changes in the testing and training set. Among these changes, misalignment of the face over testing and training effectively degrades the face recognition rate. Hence, resolving the abovementioned complications in the classical face recognition and verification technique is essential. Thus, a brand new face recognition and verification approach is developed based on optimal features and other approaches. Initially, images associated with face recognition are attained from the standard dataset and offered to pre-processing phase using median filtering. Then, the pre-processed images are fed to the feature extraction phase, the spatial features are acquired by Local Binary Pattern (LBP), and the spectral features are attained by 3 Discrete Wavelet Transform (3-DWT) levels. Later, the attained features are concatenated and utilized to obtain the optimal features by the Adaptive Crossover and Stallions Percentage-based Wild Horse Optimization (ACSP-WHO). Further, the attained optimal features are fed as the input to Ensemble Learning Network (ELNet)-based face recognition and verification phase, where the ensemble technique having Support Vector Machine (SVM), Deep Neural Networks (DNN), and Adaboost are utilized, and also the ACSP-WHO tunes their parameters to attain the practical face recognized and verified outcome in the developed model. Finally, the suggested face recognition and verification framework secured an effective efficacy rate than the traditional approaches in different experimental observations.