Variant-Age Face Recognition Using the Haar Algorithm
Open Access
- 20 October 2026
- journal article
- research article
- Published by MDPI AG in Big Data and Cognitive Computing
- Vol. 10 (10) , 357
Abstract
Cross-age face recognition remains a challenge as people’s facial appearance changes significantly over time, reducing the reliability of conventional recognition systems. The proposed framework introduces a confidence-driven age-aware matching strategy that integrates demographic filtering with adaptive similarity fusion. The proposed system employs a multi-stage architecture comprising (1) locating the face via the Haar Cascade and AdaBoost classifiers to ignore non-face areas, (2) performing age detection fusing PCA-based classification with facial feature analysis across five age groups, and (3) implementing a principled decision fusion strategy that weights age-constrained versus overall matches based on confidence estimates to produce the final match. Improvements in performance are clearly evident in the comprehensive evaluation of the FG-Net aging dataset versus the baseline PCA-based face recognition. The proposed method achieves 96.48% overall accuracy on FG-Net versus 93.66% on the baseline, whereas GIRAF remains steady without any improvement after applying age filtering, thus demonstrating an overall accuracy of 97.35% for both the baseline and proposed framework; the GIRAF dataset is presented as a supplementary evaluation and does not provide equivalent evidence regarding the benefit of age filtering. The evaluation metric of the F1-score notably improved to 0.9648 on FG-Net, indicating robust recognition across individuals regardless of sample availability. Additionally, the introduction of age filtering increases the processing time because an additional age-estimation stage is executed before identity matching. Nevertheless, the PCA representation substantially reduces the dimensionality of the original 22,500-pixel representation, thereby reducing the computational burden of the subsequent matching stage. The age-guided approach therefore reduces the standard deviation (STD), demonstrating enhanced stability, less sensitivity to aging, and robustness. Our proposed age-aware facial-recognition system offers advantages in authentication applications. Additionally, the decision-making process with interpretable confidence metrics is efficient and suitable for near-real-time applications under the tested hardware configuration. However, limitations arise from the relatively small size of the dataset and PCA sensitivity to alignment and illumination. In addition, there is a probability of incorrect discrete age categorization, which can cause misleading results. Future work will explore replacing the Haar Cascade detector and PCA with modern deep learning-based methods for face and age estimation, as well as evaluating the system on larger aging datasets such as CACD. The proposed framework demonstrates that lightweight age-aware recognition can substantially improve recognition robustness while maintaining low computational complexity.Keywords
Funding Information
- Multimedia University
This publication has 17 references indexed in Scilit:
- Face Detection and Recognition Using Face Mesh and Deep Neural NetworkProcedia Computer Science, 2023
- Facial recognition using Haar cascade and LBP classifiersGlobal Transitions Proceedings, 2021
- Prediction of face age progression with generative adversarial networksMultimedia Tools and Applications, 2021
- A survey of face recognition techniques under occlusionIET Biometrics, 2021
- A Review of Face Recognition TechnologyIEEE Access, 2020
- Face detection techniques: a reviewArtificial Intelligence Review, 2018
- A practical implementation of face detection by using Matlab cascade object detector2015
- Age‐invariant face recognition system using combined shape and texture featuresIET Biometrics, 2015
- Rapid object detection using a boosted cascade of simple features2005
- Toward automatic simulation of aging effects on face imagesIEEE Transactions on Pattern Analysis and Machine Intelligence, 2002