EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Comparative Analysis of Automatic Facial Segmentation and Cropping

Ayu Wirdiani, Agung Oka Sudana, Dwi Rusjayanthi, I Putu Ardana

International Journal on Advanced Science, Engineering and Information Technology · 11.10.2026 · 16 (5) · 1535-1542

DOI: 10.18517/ijaseit.16.5.21762 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal on Advanced Science, Engineering and Information Technology, online erschienen am 11.10.2026. Verfasst von Ayu Wirdiani und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Facial segmentation and cropping are important preprocessing steps for constructing facial datasets for biometric and facial analysis applications. Existing approaches commonly use facial landmark detection or pixel-level segmentation, but their ability to preserve important facial components may vary across different facial poses. This research addresses this gap through a qualitative comparative analysis of three representative methods, such as dlib, Face Alignment Network (FAN), and U-Net, for automatic facial segmentation and cropping of frontal, semi-side, and side-view facial images. Rather than developing a new model, this research analyzes landmark-localization consistency and the preservation of key anatomical components, including the eyebrows, eyes, nose, and mouth, in the resulting cropped regions. We evaluated Dlib and FAN using pretrained landmark detection models, while we trained U-Net on facial images from the LaPa dataset for pixel-level segmentation. We assessed landmark-based methods by the completeness and consistency of detected facial components and the success of the resulting cropping, whereas we evaluated U-Net using F1-score and Jaccard index. The results showed that FAN provided more stable landmark localization across poses, while dlib executed faster but showed more detection errors in non-frontal views. U-Net produced more detailed facial boundaries, achieving an average F1-score of 72.48% and Jaccard index of 68.75%, but required the highest computational time. These findings provide practical insights into the strengths and limitations of each method for preparing facial datasets intended for face recognition, emotion recognition, and feature-specific facial analysis.

Zitieren

APA

Wirdiani, A., Sudana, A. O., Rusjayanthi, D., & Ardana, I. P. (2026). Comparative Analysis of Automatic Facial Segmentation and Cropping. International Journal on Advanced Science, Engineering and Information Technology, 16(5), 1535-1542. https://doi.org/10.18517/ijaseit.16.5.21762

Vancouver

Wirdiani A, Sudana AO, Rusjayanthi D, Ardana IP. Comparative Analysis of Automatic Facial Segmentation and Cropping. International Journal on Advanced Science, Engineering and Information Technology. 2026;16(5):1535-1542. doi:10.18517/ijaseit.16.5.21762

BibTeX
@article{Wirdiani2026,
  author = {Wirdiani, Ayu and Sudana, Agung Oka and Rusjayanthi, Dwi and Ardana, I Putu},
  title = {Comparative Analysis of Automatic Facial Segmentation and Cropping},
  journal = {International Journal on Advanced Science, Engineering and Information Technology},
  year = {2026},
  volume = {16},
  number = {5},
  pages = {1535--1542},
  doi = {10.18517/ijaseit.16.5.21762}
}
Kostenlos starten →