Efficient large-scale face clustering using an online Mixture of Gaussians

In this work, we address the problem of large-scale online face clustering: given a continuous stream unknown faces, create database grouping incoming faces by their identity. The must be updated every time new arrives. addition, solution efficient, accurate and scalable. For purpose, present an gaussian mixture-based clustering method (OGMC). key idea is proposal that identity can represented more than just one distribution or cluster. Using feature vectors (f-vectors) extracted from OGMC generates clusters may connected to others depending on proximity robustness. Every cluster with sample, its connections are also updated. With approach, reduce dependency process order size data able deal complex distributions. Experimental results show proposed approach outperforms state-of-the-art methods benchmarks not only in accuracy, but efficiency scalability.