Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Facial expression analysis and expression-invariant face recognition by manifold-based synthesis
Peng Y., Yin H. Machine Vision and Applications29 (2):263-284,2018.Type:Article
Date Reviewed: Aug 23 2018

Although an old topic of exploration, facial expression has become an important field of research. Marketing and medical research frequently use emotion analysis to identify customer needs and emotions.

The present paper deals with facial expression, but it is about expression-invariant face recognition rather than emotion detection. The research addresses the need to supplement the training set with synthesized images of possibly missing facial expressions to augment the accuracy of the face recognition task. Moreover, the paper integrates straightforward techniques with generally acknowledged methods to demonstrate that realistic facial expressions can be generated from only one face image. This represents an important contribution of the work. Six basic expressions (anger, disgust, fear, happiness, sadness, and surprise) with five intensity levels are assumed.

Section 2 presents an interesting report on the state-of-the-art approaches to expression and face recognition.

Section 3 makes a detailed and sound account of the synthesis algorithm. The proposed approach is based on the active appearance model, in which faces are represented as weighted combinations of shape and texture attributes. The authors consider the principal component analysis (PCA) approach for expression reconstruction from eigenfaces, and derive a similar paradigm to represent neutral or expressive faces via weighted combinations of appropriate images in the training set. Specific issues are tackled with adequate approaches. Shape alignment, for instance, is made by matching eyes coordinates--a more suited solution than the standard generalized Procrustes analysis (GPA) technique, as important variations are involved. Texture matching is based on Delaunay triangulation. The transfer of expressions from source to target subjects uses differences of pairs of expressions, and Wiener filtering for noise removal. Transfer techniques and similarity measures are applied for teeth region synthesis. To include expression intensity, the authors add a dynamic synthesis scheme, expression manifold-based synthesis (EMS), which generates smooth expression manifolds through interpolation and transfer operations.

Section 4 reviews modeling and classification approaches to be applied in the evaluation stage. The authors consider expression identification and expression-invariant face recognition. Several classification methods are mentioned, among them kernel discriminant analysis, other discriminant analysis methods, and manifold-to-manifold distance.

The solution is extensively tested on a broad range of facial expression databases, including Bosphorus and the Japanese female facial expression (JAFFE) database. The tests address expression synthesis quality, expression recognition, and expression-invariant face recognition by evaluating several configurations (for instance, both including and omitting the dynamic EMS strategy) and classification schemes.

The work is very creative, original, and grounded on good knowledge of the face and image recognition fields.

Reviewer:  Svetlana Segarceanu Review #: CR146216 (1811-0598)
Bookmark and Share
 
Object Recognition (I.4.8 ... )
 
 
Eigenvalues And Eigenvectors (Direct And Iterative Methods) (G.1.3 ... )
 
 
Pattern Analysis (I.5.2 ... )
 
 
Expressions And Their Representation (I.1.1 )
 
Would you recommend this review?
yes
no
Other reviews under "Object Recognition": Date
On view likelihood and stability
Weinshall D., Werman M. IEEE Transactions on Pattern Analysis and Machine Intelligence 19(2): 97-108, 1997. Type: Article
Feb 1 1999
Junctions
Parida L., Geiger D., Hummel R. IEEE Transactions on Pattern Analysis and Machine Intelligence 20(7): 687-698, 1998. Type: Article
Apr 1 1999
Topological recognition of polyhedral objects from multiple views
Laurentini A. Artificial Intelligence 127(1): 31-55, 2001. Type: Article
Apr 17 2002
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy