Scholarlink

<p>Facial structures provide distinctive features for demographic analysis; however, variations in pose, illumination, and background remain significant obstacles to precise age and gender classification. Accurate systems are vital for social robotics, human-computer interaction, and age-restricted access control. This paper proposes a synergistic hybrid framework that integrates handcrafted texture descriptors with deep spatial features through serial fusion. Local texture features are extracted from grayscale images using an Enhanced Single-Channel Grayscale Directional Local Quinary Pattern (ESCG-DLQP) with adaptive thresholding to ensure illumination robustness. Concurrently, deep global features are extracted via a customized VGG19 backbone. Both feature sets undergo Principal Component Analysis (PCA) for dimensionality reduction before being concatenated for classification. Rigorous evaluation on the UTKFace dataset demonstrates that the fused hybrid model demonstrates competitive performance accuracies of 94.03% for age and 99.31% for gender using 70-30% split, representing a 15.68% performance uplift in age estimation over the baseline pattern. Similarly, using 5-fold cross-validation testing the proposed model achieved a mean accuracy of 90.29±0.29% for age group and 96.99±0.14% for gender classification. Furthermore, cross-dataset validation on Adience and IMDB-WIKI benchmarks confirms robust generalization, with an average performance improvement of +6.90% for age and +2.35% for gender over existing techniques. A systematic performance variance analysis highlights that the fusion of micro-patterns with semantic deep hierarchies consistently outclasses standalone handcrafted or deep-learning approaches across all standard metrics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Facial Age and Gender Classification Using Serial Fusion of ESCG-DLQP and Deep Features

  • 期刊论文

  • 发表日期：

    2026年5月30日

查看全文
  • Wakeel Ahmad,
  • Ammad Noor,
  • Syed M. Adnan,
  • Hafiz Syed Muhammad Bilal

摘要

Facial structures provide distinctive features for demographic analysis; however, variations in pose, illumination, and background remain significant obstacles to precise age and gender classification. Accurate systems are vital for social robotics, human-computer interaction, and age-restricted access control. This paper proposes a synergistic hybrid framework that integrates handcrafted texture descriptors with deep spatial features through serial fusion. Local texture features are extracted from grayscale images using an Enhanced Single-Channel Grayscale Directional Local Quinary Pattern (ESCG-DLQP) with adaptive thresholding to ensure illumination robustness. Concurrently, deep global features are extracted via a customized VGG19 backbone. Both feature sets undergo Principal Component Analysis (PCA) for dimensionality reduction before being concatenated for classification. Rigorous evaluation on the UTKFace dataset demonstrates that the fused hybrid model demonstrates competitive performance accuracies of 94.03% for age and 99.31% for gender using 70-30% split, representing a 15.68% performance uplift in age estimation over the baseline pattern. Similarly, using 5-fold cross-validation testing the proposed model achieved a mean accuracy of 90.29±0.29% for age group and 96.99±0.14% for gender classification. Furthermore, cross-dataset validation on Adience and IMDB-WIKI benchmarks confirms robust generalization, with an average performance improvement of +6.90% for age and +2.35% for gender over existing techniques. A systematic performance variance analysis highlights that the fusion of micro-patterns with semantic deep hierarchies consistently outclasses standalone handcrafted or deep-learning approaches across all standard metrics.