Globally Maximizing, Locally Minimizing: Unsupervised Discriminant Projection with Applications to Face and Palm Biometrics
read more
Citations
Maximum neighborhood margin discriminant projection for classification
Extended SRC: Undersampled Face Recognition via Intraclass Variant Dictionary
A survey of palmprint recognition
Gabor feature based sparse representation for face recognition with gabor occlusion dictionary
Fusing Local Patterns of Gabor Magnitude and Phase for Face Recognition
References
Nonlinear dimensionality reduction by locally linear embedding.
Eigenfaces for recognition
A global geometric framework for nonlinear dimensionality reduction.
Eigenfaces vs. Fisherfaces: recognition using class specific linear projection
Nonlinear component analysis as a kernel eigenvalue problem
Related Papers (5)
Nonlinear dimensionality reduction by locally linear embedding.
A global geometric framework for nonlinear dimensionality reduction.
Frequently Asked Questions (13)
Q2. What are the well-known methods for finding intrinsic low-dimensional embedding of data?
Among the most well-known are isometric feature mapping (ISOMAP) [22], local linear embedding (LLE) [23], and Laplacian Eigenmap [24].
Q3. Why are manifold learning algorithms unsuitable for pattern recognition tasks?
The second reason why most manifold learning algorithms, for example, ISOMAP, LLE, and Laplacian Eigenmap, are unsuitable for pattern recognition tasks is that they can yield an embedding directly based on the training data set but, because of the implicitness of the nonlinear map, when applied to a new sample, they cannot find the sample’s image in the embedding space.
Q4. What is the locality radius of the ellipse?
1. If the locality radius is set as the length of the semimajor axis of the larger ellipse, the direction w1 is a nice projection according to the criterion of LPP since, after all samples are projected onto w1, the local scatter is minimal.
Q5. What is the reason why the authors use only the cosine distance metric?
Since the cosine distance is more effective than the Euclidean distance for LDA, Laplacianface, and UDP, in the following experiments the authors use only this distance metric.
Q6. What is the way to achieve an optimal recognition result?
For achieving an optimal recognition result, the recovered embeddings corresponding to different face manifolds should be as separate as possible in the final embedding space.
Q7. What is the way to find the optimal projection for clustering in the observed space?
Provided that each cluster of samples in the observation space is exactly within a local neighbor, UDP can yield an optimal projection for clustering in the projected space, while LPP cannot.
Q8. What is the definition of a linear approximation of the nonlinear map?
The projection of UDP can be viewed as a linear approximation of the nonlinear map that uncovers and separates embeddings corresponding to different manifolds in the final embedding space.
Q9. How many images are used in the training sample set?
In their experiments, l images (l varies from 2 to 6) are randomly selected from the image gallery of each individual to form the training sample set.
Q10. What is the way to improve the performance of PCA?
the cosine distancemetric can significantly improve the performance of LDA,Laplacianface, and UDP, but it has no substantial effect on theperformance of PCA.
Q11. What is the criterion for the projection of samples?
LDA seeks to find a projection axis such that the Fisher criterion (i.e., the ratio of the between-class scatter to the within-class scatter) is maximized after the projection of samples.
Q12. What are the two popular nonlinear dimensionality reduction techniques?
A number of nonlinear dimensionality reduction techniques have been developed to address this problem, with two in particular attracting wide attention: kernel-based techniques and manifold learningbased techniques.
Q13. What is the subset of images that are marked with two-character strings?
It is composed of the images whose names aremarked with two-character strings: “ba,” “bj,” “bk,” “be,” “bf.”This subset involves variations in facial expression, illumina-tion, and pose.