Efficient Gender Recognition with Deep Learning Models on Balanced Facial Image Datasets
Kishan Poddar, Bhanu Pratap Sharma · 2024
From security to marketing to human-computer interaction, gender recognition from facial photographs is a basic task in many disciplines. This effort intends to develop an appropriate gender recognition model using a balanced dataset including 1,292 images equally distributed between men and women. Training, validation and testing sets were created from diverse age groups, ethnicities and environments of the dataset to ensure proper evaluation. Transfer learning methods were used to accelerate training and increase generalization using ResNet50, a deep learning model developed on convolutional neural networks (CNN). Designed for the specific difficulty of gender recognition, the pre-trained model was developed on leveraging characteristics obtained from a large dataset to achieve amazing accuracy with minimal data and processing resources. Both suitable for classification issues, the model was optimized using cross-entropy loss and Adam optimizer. Performance evaluation was shown by remarkable accuracy of 97% coupled with outstanding precision, recall and F1-scores including both genders. Low rates of misclassification help to establish even more the dependability of the model by use of the confusion matrix. These results show the viability of the approach and its use in pragmatic gender recognition programs.