Niqab Dataset for Occluded Face Detection

The Niqab Dataset is a curated collection of images designed to support research in face detection under conditions of heavy occlusion, particularly relevant for cultural and religious coverings such as the niqab, veil, and mask. The dataset contains approximately 10,000 images with 12,000 faces, where over 50% of the faces are covered. A subset of 5,175 images (5,343 faces) was employed in our experiments.

The dataset provides diverse occlusion styles of niqabs and veils, reflecting real-world variations in coverage, position, and texture. Each image is manually annotated with bounding boxes using a contextual labeling approach that incorporates surrounding head and shoulder regions, enabling models to capture auxiliary visual cues critical for detecting highly occluded faces.

This dataset was developed and utilized in the study:Alashbi AAS, Sunar MS. Occluded face detection, face in Niqab dataset. In: Saeed F, Mohammed F, Gazem N,editors. Emerging Trends in Intelligent Computing and Informatics. Cham, Switerland: Springer InternationalPublishing; 2020. p. 209–15. doi: 10.1007/978-3-030-33582-3_20. 

The dataset can support a wide range of applications in computer vision, security, surveillance, biometrics, and cultural AI research, serving as a benchmark for the development and evaluation of models handling highly occluded face scenarios.

Contents:

10,000+ images (niqab-covered faces, partially occluded faces, and non-faces).

Bounding box annotations (contextual labeling).

Training/testing split (approx. 80/20).

Intended Use:For research in face detection, occluded face recognition, and machine learning model benchmarking.