Facial expression recognition using clustering discriminant Non-negative Matrix Factorization
read more
Citations
Automatic Analysis of Facial Affect: A Survey of Registration, Representation, and Recognition
Projected gradients for subclass discriminant nonnegative subspace learning.
Projective complex matrix factorization for facial expression recognition
A Sparse Corruption Non-Negative Matrix Factorization method and application in face image processing & recognition
Sparse localized facial motion dictionary learning for facial expression recognition
References
Eigenfaces vs. Fisherfaces: recognition using class specific linear projection
Learning the parts of objects by non-negative matrix factorization
Learning parts of objects by non-negative matrix factorization
Algorithms for Non-negative Matrix Factorization
Related Papers (5)
Learning the parts of objects by non-negative matrix factorization
Frequently Asked Questions (14)
Q2. What is the purpose of the SDNMF algorithm?
The SDNMF algorithm addresses the general problem of finding discriminant projections that enhance class separability in the reduced dimensionality projection space.
Q3. What is the rationale to desire to have a constrained optimization problem in n-?
The constrained optimization problem in (9) is solved by introducing Lagrangian multipliers φ = [φi,k] ∈ RF×M and ψ = [ψj,k] ∈ RM×L each one associated with constraints zi,k ≥ 0 and hk,j ≥ 0, respectively.
Q4. What is the purpose of the NMF algorithm?
Focusing on the application of the NMF algorithm on facial image data, NMF aims to approximate a facial image by a linear combination of elements the so called basis images, that correspond to facial parts.
Q5. What is the main disadvantage of the DNMF?
Fisher’s criterion in the NMF decomposition and achieves a more efficient decomposition of the provided data to its discriminant parts, thus enhancing separability between classes compared with conventional NMF.
Q6. What is the simplest way to approximate facial images?
NMF considers factorizations of the form:X ≈ ZH (1)where Z ∈ RF×M+ is a matrix containing the basis images, while matrix H ∈ RM×L+ contains the coefficients of the linear combinations of the basis images required to reconstruct each original facial image in the database.
Q7. What is the main disadvantage of LDA?
LDA assumes that the sample vectors of the classes are generated from underlying multivariate Normal distributions of common covariance matrix but with different means.
Q8. What is the rationale for the dispersion of those samples that belong to the same class?
It is rationale to desire the dispersion of those samples that belong to the same cluster of a certain class to be as small as possible, since this would denote a high concentration of these samples around their cluster mean and consequently, more compact clusters formation.
Q9. What is the name of the research that led to these results?
The research leading to these results has received funding from the European Community’s Seventh Framework Programme (FP7/2007-2013) under grant agreement no 211471 (i3DPost).
Q10. What is the non negative constraint in NMF?
The non negativity constraint imposed has been exploited by a variety of applications since many types of data in practical problems are non negative.
Q11. What is the cost of factorizing X into ZH?
the cost for factorizing X into ZH is evaluated as:DNMF (X||ZH) = L ∑j=1KL(xj ||Zhj) ==L ∑j=1F ∑i=1(xi,j ln( xi,j ∑k zi,khk,j ) +∑kzi,khk,j − xi,j ) .
Q12. What is the cost of the decomposition in (1)?
To measure the cost of the decomposition in (1), one popular approach is to use the Kullback-Leibler (KL) divergence metric [7, 8].
Q13. What is the definition of a supervised NMF algorithm?
A supervised NMF learning method that aims to extract discriminant facial parts, is the Discriminant NMF (DNMF) algorithm proposed in [4].
Q14. What is the main limitation of the NMF?
To remedy the aforementioned limitations the authors relax the assumption that each class is expected to consist of a single compact cluster and regard that data inside each class form various clusters, where each one is approximated by a Gaussian distribution.