International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 05 | May 2025
p-ISSN: 2395-0072
www.irjet.net
Eye-Region-Based Gender Classification Using CNNs Battula Balnarsaiah1, Paduru Bimbasri2, Paluri Sai Varun Raj3, Nitesh Gaikwad4 1Associate Professor, Dept. Artificial Intelligence and Machine Learning, Nalla Malla Reddy Engineering College,
Hyderabad, India
2 Dept. Artificial Intelligence and Machine Learning, Nalla Malla Reddy Engineering College, Hyderabad, India 3 Dept. Artificial Intelligence and Machine Learning, Nalla Malla Reddy Engineering College, Hyderabad, India 4 Assistant Professor, Dept. Electronics and Communication Engineering, Nalla Narsimha Reddy Group Of
Institutions, Hyderabad, India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Over the last ten years, automated gender
techniques have also been increasingly applied to facial gender classification.
identification using facial images has emerged as a significant area of research, demanding effective strategies for both feature extraction and classification. Accurate feature extraction is necessary in traditional machine learning methods, while Convolutional Neural Networks (CNNs), a type of deep learning model, perform this directly using raw data. CNNs can accommodate variations in facial cues across different races, making them efficient for gender classification. However, the performance of pretrained CNN models is investigated in low-data situations. Methodology involves library imports, data organization, distribution exploration, and image visualization. Then comes label encoding and implementation of various CNN architectures, as well as performance metrics and predictions, 92 % for training 8,643 images and testing 2,882 images, and it requires only four folds. The study notably deviates from using complete face images, focusing on areas with eyebrows around one eye only to identify the gender.
2. LITERATURE REVIEW This study narrows the scope of gender classification from full facial images to focusing solely on the eye region. As a result, both the dataset and the area of interest are more specifically defined. Significant progress has been made in classifying gender from eye images, especially with the development of deep learning and neural networks. Early work on gender classification, such as the research by Jabber and Hashim et al (2018) [1], focused on robust eye feature extraction using eye angles to enhance gaze classification efficiency. These initial approaches relied heavily on manually extracted features and traditional machine learning techniques. For example, Lian HC et al [2] achieved 94.81% accuracy by employing Local Binary Patterns (LBP) in combination with Support Vector Machines (SVM). In contrast, challenges with background complexity affected their accuracy rates. With the advent of DL, researchers began leveraging CNNs and other advanced Neural Network (NN) architectures for more accurate gender classification from eye images. Notable studies include those that utilized CNN-based models. CIMTAY and YILMAZ et al (2021) [3] used pretrained CNNs to classify gender from eye images, demonstrating significant improvements in accuracy, although specific percentages were not reported.
Key Words: Gender Classification, Convolutional Neural Network, Facial Cues.
1. INTRODUCTION Artificial intelligence has become instrumental in addressing various challenges associated with human recognition. Biometric data such as facial features, fingerprints, voice, and iris patterns are commonly utilized by AI systems. Among these, gender recognition based on facial imagery remains a complex task, primarily due to the dynamic nature of human facial characteristics. Variations such as facial hair, aging, and changes in head orientation can significantly affect model accuracy. To address these challenges, researchers have suggested numerous models and methods intended to enhance the reliability and achieving high-performance outcomes.
Sri and Gupta et al (2022) [4] achieved notable accuracies by focusing on the morphometry of eyes using DL models. Advanced NN architectures have further pushed the boundaries of this research. Huang et al. (2017) developed Whole-Component CNNS (WC-CNN) [5] for age and gender classification, achieving a high accuracy rate, though specific figures were not detailed. Keshaveni et al (2024) [6] introduced an innovative approach combining AVOA-LSTM and Mask R-CNN for segmenting and classifying the sunglass image-based eye region, achieving accuracies that significantly surpassed traditional methods. Another significant advancement is the incorporation of attention mechanisms and key region fusion. Kong et al. (2021) [7] proposed a lightweight facial expression recognition
Deep Learning (DL) has received substantial attention for its capabilities in automatic feature extraction, object detection, and classification tasks. Deep Convolutional Neural Networks (DCNNs), one example of an advanced design, have shown impressive accuracy in these domains. Inspired by developments across multiple fields, deep learning
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 769