[Paper Reding] Scalable Face Image Retrieval using Attribute-Enhanced Sparse Codewords

Problem Definition

• Retrieve face images based on content from a large-scale database
• Utilize automatically detected human attributes

Contributions

• Combine automatically detected high-level human attributes and low-level features to construct semantic codewords
• Propose two orthogonal methods to utilize automatically detected
  human attributes to improve content-based face image retrieval under a scalable framework
• Experiments on public datasets

Related Works

Content-based image retrieval (CBIR)

• Color, texture and gradient
• Inverted indexing, hash-based indexing, combined with bag-of-word model (BoW) and local features like SIFT
• Semantic image representations: extra textual information, class
  labels

OBSERVATIONS

• Prior works usually crop only the facial region and normalize the face into the same position and illumination to reduce intra-class variance caused by poses and lighting variations.
• They ignore the rich semantic cues for
  a designated face such as skin color, gender, hair style.

• Hypothesize that using human attributes can help the face retrieval task

Method

System overview

Attribute-enhanced sparse coding (ASC)

1. Sparse coding for face image retrieval (SC)
2. Attribute-enhanced sparse coding (ASC)

Attribute Embedded Inverted Indexing (AEI)

1. Image ranking and inverted indexing
2. Attribute-embedded inverted indexing

Results

Reference

Chen, Bor-Chun, et al. “Scalable face image retrieval using attribute-enhanced sparse codewords.” IEEE Transactions on Multimedia 15.5 (2013): 1163–1173.