Categorization of Celebrity Photos Based on Deep Machine Learning for Feature Extraction and Classification

• Published in: Wasit Journal of Computer and Mathematics Science Vol. 4; no. 1; pp. 1 - 16
• Main Authors: Shakir Baawi, Salwa, Jawad Al-Ghanim, Farah, Ryadh Hamza, Nisreen
• Format: Journal Article
• Language: English
• Published: College of Computer and Information Technology – University of Wasit, Iraq 21.05.2025
• ISSN: 2788-5879; 2788-5887
• DOI: 10.31185/wjcms.341

In feature extraction, images or videos are analyzed to identify facial features before pinpointing the exact target. Significant progress has been made in the realm of feature extraction through deep learning. With deep learning technologies, developers have created algorithms for facial analysis and recognition that enhance accuracy and effectiveness. Consequently, programmers and developers have started to implement facial recognition technology in a wide range of applications to address these challenges. The identification process is particularly difficult due to the abundance of unstructured datasets. In real-world scenarios, identification continues to be quite challenging despite various methods proposed so far. This paper aims to analyze celebrity categorization based on deep features under different conditions. Most algorithms are influenced by these problematic factors, whether intentionally or unintentionally. Six datasets are utilized in this paper: the Open Famous People Dataset, Celeb Identification Dataset, Bollywood Celebrity Localized Dataset (170), Football Player's Dataset, Mini_LFW, and Nepali Celeb Localized Dataset. Most of these images feature celebrities including actors, singers, players, actresses, politicians, and social figures from diverse nationalities and countries. Therefore, this paper offers a combination of results from classification methods: Multilayer Perceptron Classifier (MLP), Decision Tree Classifier (DT), Bootstrap Aggregation Decision Tree Classifier (TreeBagger), Extreme Gradient Boosting Classifier (XGB), and Stochastic Gradient Descent Classification (SGD), alongside deep learning methods for feature extraction (VGGFace2 model pretraining and SENet-50 model architecture) across six different facial datasets. The results demonstrated that the highest accuracy (F-measure, Recall, and Precision) was achieved following the training and testing process in all datasets using the MLP and SGD classifiers.