Enhancing Age and Gender Detection Performance in CNNs Through Optimization of Batch Normalization and Dropout Settings with Iterative Parameter

Abstract

Facial age and gender recognition from images facilitates intelligent applications across various fields. Although convolutional neural networks (CNNs) represent the leading approach, achieving peak performance necessitates careful adjustment of regularization strategies. Building on the previous study, this research has adjusted the model by adding an iterative parameter to process batch normalization and dropout, optimizing the weights during model execution. Compared to the results of the previous study, the proposed model 2 achieved improved accuracy over proposed model 1 for gender and age detection, with improvements of 3% and 2%, respectively. These findings underscore the importance of fine-tuning batch normalization and dropout for developing robust CNNs with loop parameter. Our research offers valuable perspectives on optimizing CNN architecture and regularization in the field of computer vision.