A portable clustering algorithm based on compact neighbors for face tagging.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Facial image analysis often requires grouping images by individual identity.
• Clustering unlabeled facial data presents challenges, especially when the number of subjects is unknown.
• Existing methods may be sensitive to parameter choices or computationally intensive.

Purpose of the Study:

• To propose a novel algorithm for unsupervised face clustering.
• To develop a method that is robust to parameter selection and computationally efficient.
• To address the challenge of unknown subject numbers in facial image collections.

Main Methods:

• Development of the Portable Clustering algorithm based on Compact Neighbors (PCN).
• Utilizing compact neighbors to automatically determine local sample density.
• Employing a two-stage framework: deep feature extraction followed by clustering in feature space.
• Experimental validation on 16 middle- and 5 large-scale benchmark datasets.

Main Results:

• PCN automatically determines local density with only one user-specified parameter (k-nearest neighbors).
• The algorithm's performance is robust and not sensitive to the choice of k.
• PCN exhibits near-linear computational complexity, ensuring scalability to large datasets.
• Experimental results confirm PCN's superior efficiency and effectiveness compared to state-of-the-art methods.

Conclusions:

• PCN offers an effective and scalable solution for unsupervised face clustering.
• The algorithm's robustness and efficiency make it suitable for practical applications.
• PCN advances the state-of-the-art in facial image analysis and clustering techniques.