EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Multi-Descriptor Fusion with Deep Residual Learning for Kinshipship Identification from Facial Images

Munzza Bibi, Wakeel Ahmad, Syed M. Adnan, Asifa Bibi

Journal of Computing Theories and Applications · 18.08.2026 · 4 (1) · 369-386

DOI: 10.62411/jcta.16066 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Computing Theories and Applications, online erschienen am 18.08.2026. Verfasst von Munzza Bibi und 3 weiteren Autor:innen. Die Arbeit nennt 45 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Kinship identification from facial images aims to determine biological relationships between individuals based on shared facial characteristics. However, subtle kinship-related facial cues are often obscured by variations in illumination, pose, age, and facial expressions, making reliable kinship classification challenging. To address this problem, this study proposes a hybrid framework that integrates handcrafted texture descriptors with deep features for multiclass kinship identification. A preprocessing pipeline consisting of image resizing to 224 × 224 pixels, noise reduction, intensity normalization, and facial region extraction is first applied to improve image consistency and feature quality. Three complementary local texture descriptors, namely Local Binary Pattern (LBP), Local Ternary Pattern (LTP), and Local Directional Pattern (LDP), are then employed to capture fine-grained facial texture and directional information. These handcrafted representations are fused with 2048-dimensional deep features extracted from a ResNet-50 model. The resulting 4562-dimensional pair representation is classified using a Support Vector Machine (SVM) under a four-class setting comprising father–son, father–daughter, mother–son, and mother–daughter relationships. Experiments on the KinFaceW-I dataset demonstrate that the proposed hybrid framework achieves a mean accuracy of 82.97%, with mean precision, recall, and F1-score of 82.99%, 83.00%, and 82.98%, respectively. The results further show consistent performance across all four relationship categories and competitive performance against existing methods, demonstrating the effectiveness of combining complementary local texture descriptors with deep semantic representations.

Zitieren

APA

Bibi, M., Ahmad, W., Adnan, S. M., & Bibi, A. (2026). Multi-Descriptor Fusion with Deep Residual Learning for Kinshipship Identification from Facial Images. Journal of Computing Theories and Applications, 4(1), 369-386. https://doi.org/10.62411/jcta.16066

Vancouver

Bibi M, Ahmad W, Adnan SM, Bibi A. Multi-Descriptor Fusion with Deep Residual Learning for Kinshipship Identification from Facial Images. Journal of Computing Theories and Applications. 2026;4(1):369-386. doi:10.62411/jcta.16066

BibTeX
@article{Bibi2026,
  author = {Bibi, Munzza and Ahmad, Wakeel and Adnan, Syed M. and Bibi, Asifa},
  title = {Multi-Descriptor Fusion with Deep Residual Learning for Kinshipship Identification from Facial Images},
  journal = {Journal of Computing Theories and Applications},
  year = {2026},
  volume = {4},
  number = {1},
  pages = {369--386},
  doi = {10.62411/jcta.16066}
}
Kostenlos starten →