Generative face inpainting hashing for occluded face retrieval

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• The COVID-19 pandemic created unprecedented challenges for face recognition systems, particularly due to widespread face mask usage causing occlusions.
• Accurate face retrieval from occluded images is crucial for various applications, including security and identification.

Purpose of the Study:

• To develop an effective method for face retrieval in large-scale datasets where faces are subject to various types of occlusion.
• To improve the performance of deep hashing retrieval techniques for occluded face images.

Main Methods:

• A novel occluded face retrieval method combining a face inpainting model with a deep hashing retrieval network.
• The face inpainting model is trained using adversarial loss, reconstruction loss, and hash bits loss to reconstruct occluded face images.
• A deep hashing retrieval network generates compact, similarity-preserving hashing codes from reconstructed images.

Main Results:

• The proposed method successfully reconstructs occluded face images.
• The deep hashing retrieval network demonstrates superior retrieval performance for occluded faces compared to existing state-of-the-art methods.
• Experimental results validate the effectiveness of the proposed approach in handling diverse occlusion scenarios.

Conclusions:

• The developed method offers a robust solution for face retrieval under occlusion, addressing a significant challenge exacerbated by the COVID-19 pandemic.
• The integration of face inpainting and deep hashing significantly enhances retrieval accuracy for occluded faces.