Eigenspace-based face recognition: a comparative study of different approaches
read more
Citations
A Comparative Study of Local Matching Approach for Face Recognition
Face recognition using HOG-EBGM
Incremental Linear Discriminant Analysis for Face Recognition
Recognition of faces in unconstrained environments: a comparative study
Studies on Hyperspectral Face Recognition in Visible Spectrum With Feature Band Selection
References
Support-Vector Networks
A Tutorial on Support Vector Machines for Pattern Recognition
Eigenfaces for recognition
Related Papers (5)
Frequently Asked Questions (11)
Q2. What have the authors stated for future works in "Eigenspace-based face recognition: a comparative study of different hybrid approaches" ?
As future work the authors would like to extend their study by considering other kernel approaches and algorithms, as for example ICA ( Independent Component Analysis ), Kernel-ICA, and new algorithms as FLDA ( Fractional-step Linear Discriminant Analysis ) [ 6 ] and DF-LDA ( Direct F-LDA ) [ 7 ] that improve standard FLD.
Q3. What are the main problems of kernel methods?
Although kernel methods obtain the best recognition rates, they suffer from problems such as lowprocessing speed and the difficulty to adjust the kernel parameters.
Q4. What is the main objective of a similarity measure?
The main objective of a similarity measure is to define a value that allows the comparison of feature vectors (reduced vectors in eigenspace frameworks).
Q5. What is the main idea behind the eigenspace-based methods?
Eigenspace-based methods, mostly derived from the Eigenface-algorithm [19], project input faces onto a dimensional reduced space where the recognition is carried out, performing a holistic analysis of the faces.
Q6. Why does the differential approach outperform the standard methods?
The reason seems to be the better generalization ability of the differential approaches, derived from the fact that more data is available for training, because by using differential images or differential vectors the recognition task becomes a two-class problem.
Q7. What was the first eigenspace-based face recognition approach?
in 1991 Turk and Pentland used PCA projections as the feature vectors to solve the problem of face recognition, using the Euclidean distance as similarity function [19].
Q8. What are the main issues that should be considered?
Other issues that should also be considered:- Post-differential approaches are 2 to 5 times faster than the pre-differential ones.
Q9. What is the reason for the better generalization ability of KPCA over KFD?
The reason seems to be the better generalization ability of KPCA over KFD, which is tested when the identification of either 127 o 254 classes is solved using just 2 training images per class.
Q10. What are the main characteristics of the two types of databases used for performing the study?
It is important to use both kinds of databases for performing such a study, because, as it will be shown in this work, some properties of the methods, as for example their generalization ability, change depending on the number of classes taken under consideration.
Q11. How many dimensions can be generated from low-resolution face images?
Even low-resolution face images generate huge dimensional feature spaces (20,000 dimensions in the case of a 100x200 pixels face image).