Bald eagle search optimization with deep transfer learning enabled age-invariant face recognition model

Abstract

Facial aging variation is a challenging process in the design of face recognition system because of high intra-personal differences instigated by age progression. Age-invariant face recognition (AIFR) models find applicability in several real time applications such as criminal identification, missing person detection, and so on. The main issue is the high intra-personal disparities because of complicated and non-linear age progression process. An essential component of face recognition model is the extraction of important features from the facial images for reducing intrapersonal differences produced by illumination, expression, pose, age, etc. The recent advances of machine learning (ML) and deep learning (DL) models pave a way for effective design of AIFR models. In this view, this study presents a new Bald Eagle Search Optimization with Deep Transfer Learning Enabled AFIR (BESDTL-AIFR) model. The presented BESDTL-AIFR model primarily pre-processes the facial images to enhance the quality. Besides, the BESDTL-AIFR model utilizes Inception v3 model for learning high level deep features. Next, these features are passed into the optimal deep belief network (DBN) model for face recognition. Finally, the hyperparameters of the DBN model are optimally chosen by the use of BES algorithm. Experimentation analysis on challenging benchmark datasets pointed out the promising outcomes of the BESDTL-AIFR model compared to recent approaches. • Develop a deep learning based Age-Invariant Face Recognition model. • Employ Inception v3 feature extractor with DBN classification model. • Present bald eagle search optimization based hyperparameter tuning process. • Validate the performance of BESDTL-AIFR model on UTKFace and CACD dataset.