A Hybrid Deep–Handcrafted Feature Fusion Framework with Attention-Weighted Ensemble Classification for Robust Face Recognition
Journal of Intelligent Decision Making and Information Science · 23.09.2026 · 3 (14s) · 1568-1602
DOI: 10.17762/jidmis.v3.4242 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Intelligent Decision Making and Information Science, online erschienen am 23.09.2026. Verfasst von Avishek Dey. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Face recognition remains one of the most extensively studied problems in computer vision because unconstrained acquisition conditions such as pose variation, non-uniform illumination, partial occlusion and expression change continue to degrade the reliability of automated identification systems. This paper proposes a hybrid recognition framework that fuses deep convolutional neural network (CNN) embeddings with handcrafted Local Binary Pattern (LBP) and Histogram of Oriented Gradients (HOG) descriptors through an attention-weighted feature-fusion layer, and classifies the fused representation using a heterogeneous ensemble composed of a Support Vector Machine (SVM), a Random Forest (RF) and a softmax neural head combined by weighted soft voting. An adaptive pre-processing pipeline — resizing, Gaussian denoising, Contrast-Limited Adaptive Histogram Equalization (CLAHE), Haar-cascade face detection and landmark-guided alignment — normalises raw images prior to feature extraction. The framework is developed and evaluated on a face-recognition benchmark assembled from publicly available Kaggle and GitHub repositories (described in Section 3). Experimental results obtained through stratified 5-fold cross-validation indicate that the proposed hybrid model attains an average recognition accuracy of 97.1%, outperforming Eigenfaces (78.4%), Fisherfaces (81.2%), Local Binary Pattern Histogram-LBPH (85.6%), an SVM+HOG baseline (87.9%) and a CNN-only baseline (93.4%). A paired t-test confirms that the improvement over the strongest baseline is statistically significant (p < 0.01), supporting the research hypothesis that attention-weighted fusion of deep and handcrafted descriptors yields a measurable accuracy gain over either representation used alone, particularly under pose and illumination variation. An ablation study quantifies the individual contribution of each architectural component, and a complexity analysis shows the proposed system maintains a practical inference latency suitable for near-real-time deployment. The reported methodology, validation protocol and quantitative results are intended as a reproducible template; readers who adopt this framework for a specific deployment should retrain and re-validate on their own captured data before operational use.
Zitieren
APA
Dey, A. (2026). A Hybrid Deep–Handcrafted Feature Fusion Framework with Attention-Weighted Ensemble Classification for Robust Face Recognition. Journal of Intelligent Decision Making and Information Science, 3(14s), 1568-1602. https://doi.org/10.17762/jidmis.v3.4242
Vancouver
Dey A. A Hybrid Deep–Handcrafted Feature Fusion Framework with Attention-Weighted Ensemble Classification for Robust Face Recognition. Journal of Intelligent Decision Making and Information Science. 2026;3(14s):1568-1602. doi:10.17762/jidmis.v3.4242
BibTeX
@article{Dey2026,
author = {Dey, Avishek},
title = {A Hybrid Deep–Handcrafted Feature Fusion Framework with Attention-Weighted Ensemble Classification for Robust Face Recognition},
journal = {Journal of Intelligent Decision Making and Information Science},
year = {2026},
volume = {3},
number = {14s},
pages = {1568--1602},
doi = {10.17762/jidmis.v3.4242}
}
EUVIMED