Parts-based approaches may generalise better on unseen data. They can also help to work around issues such as occlusion or when the shapes are not rigid (like expression on a face). Or sometimes you may have a reason to assume there is a natural cause to why your data should be defined from parts. Non-rigidity, by the way, can to some extent be solved by using a shape model rather than the raw image data.

There are feature point descriptors other than SIFT specifically developed and possibly better suited for 3D data. I don't have much experience with that though.

Another idea perhaps may be to use other matrix decompositions than PCA that lead to a parts-based approach. For example nonnegative matrix factorisation (NMF) gives a sparse basis, where the positive elements in each basis vector are found to form local patches on the face.

While at that, you might want to consider FisherFace approach too

ftp://ftp.idiap.ch/pub/courses/EE-700/material/17-10-2012/fisherface-pami97.pdf

Parts-based approaches may generalise better on unseen data. They can also help to work around issues such as occlusion or when the shapes are not rigid (like expression on a face). Or sometimes you may have a reason to assume there is a natural cause to why your data should be defined from parts. Non-rigidity, by the way, can to some extent be solved by using a shape model rather than the raw image data.

There are feature point descriptors other than SIFT specifically developed and possibly better suited for 3D data. I don't have much experience with that though.

Another idea perhaps may be to use other matrix decompositions than PCA that lead to a parts-based approach. For example nonnegative matrix factorisation (NMF) gives a sparse basis, where the positive elements in each basis vector are found to form local patches on the face.

I have used wavelets for features extraction and obtained very good results

Nowadays, holistic descriptors are commonly used in several visual recognition topics, achieving great performance results. Furthermore, the application of global image descriptors based on local binary descriptors such as BRIEF-Gist [1] or ABLE+LDB [2] has been succesfully tested in place recognition and, perhaps, it could be easily adapted to face recognition.

[1] N. Sunderhauf and P. Protzel, “BRIEF-Gist - Closing the loop by simple means”, in

IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), San Francisco (United States), September 2011, pp. 1234–1241.

[2] R. Arroyo, P. F. Alcantarilla, L. M. Bergasa, J. J. Yebes and S. Gámez, "Bidirectional Loop Closure Detection on Panoramas for Visual Navigation", in IEEE Intelligent Vehicles Symposium (IV), Dearborn, Michigan (United States), June 2014.

Conference Paper Bidirectional Loop Closure Detection on Panoramas for Visual...

Good way to analyse what approaches is good -- look at the results in modern database for face recognition. I suggest LFW (Labeled Faces in the Wild) http://vis-www.cs.umass.edu/lfw/results.html.

1. Best resaults is using convolution neural network by facebook [2] and face++[1].

2. Another good point is to use key point -- not blobs or coners -- special facial keypoints. I recomend use IntraFace

3. Combination of local features such as LBP, HOG (SIFT), gabor with LDA (or imroving of LDA)-- leads to good results [3]. This easeast way to achieve good result.

[1] Fan H. et al. Learning Deep Face Representation //arXiv preprint arXiv:1403.2802. – 2014.

[2] Yaniv Taigman, Ming Yang, Marc'Aurelio Ranzato, Lior Wolf.

DeepFace: Closing the Gap to Human-Level Performance in Face Verification.

[3] Dong Chen, Xudong Cao, Fang Wen, and Jian Sun. Blessing of Dimensionality: High-dimensional Feature and Its Efficient Compression for Face Verification.

Hello Angel, Please go through these point which are to be concentrate

Feature-based (structural) Methods:

In this methods local features such as eyes, nose and mouth are first of all extracted and their locations and local statistics (geometric and/or appearance) are fed into a structural classifier. A big challenge for feature extraction methods is feature "restoration", this is when the system tries to retrieve features that are invisible due to large variations, e.g. head Pose when

A frontal image with a profile image.

Distinguishes between three different extraction methods:

I. Generic methods based on edges, lines, and curves

II. Feature-template-based methods

III. Structural matching methods that take into consideration geometrical Constraints on the

features.

Hybrid Methods:

Hybrid face recognition systems use a combination of both holistic and feature extraction methods. Generally 3D Images are used in hybrid methods. The image of a person's face is caught in 3D, allowing the system to note the curves of the eye sockets, for example, or the shapes of the chin or forehead. Even a face in profile would serve because the system uses depth, and an axis of measurement, which gives it enough information to construct a full face. The 3D system usually proceeds thus: Detection, Position, Measurement, Representation and Matching. Detection- Capturing a face either a scanning a photograph or photographing a person's face in real time. Position- Determining the location, size and angle of the head.

Measurement- Assigning measurements to each curve of the face to make a template with specific focus on the outside of the eye, the inside of the eye and the angle of the nose. Representation- Converting the template into a code - a numerical representation of the face and Matching- Comparing the received data with faces in the existing database.

In Case the 3D image is to be compared with an existing 3D image, it needs to have no alterations. Typically, however, photos that are put in 2D, and in that case, the 3D image need a few changes. This is tricky, and is one of the biggest challenges in the field today.

Face Recognition Applications

Face recognition is also useful in human computer interaction, virtual reality, database recovery, multimedia, computer entertainment, information security e.g. operating system, medical records, online banking., Biometric e.g. Personal Identification - Passports, driver licenses , Automated identity verification - border controls , Law enforcement e.g. video surveillances , investigation , Personal Security - driver monitoring system, home video surveillance system.

Applications & examples:

Face Identification: Face recognition systems identify people by their face images. Face recognition systems establish the presence of an authorized person rather than just checking whether a valid identification (ID) or key is being used or whether the user knows the secret personal identification numbers (Pins) or passwords. The following are example.

To eliminate duplicates in a nationwide voter registration system because there are cases where the same person was assigned more than one identification number. The face recognition system directly compares the face images of the voters and does not use ID numbers to differentiate one from the others. When the top two matched faces are highly similar to the query face image, manual review is required to make sure they are indeed different persons so as to eliminate duplicates.

Access Control: In many of the access control applications, such as office access or computer logon, the size of the group of people that need to be recognized is relatively small. The face pictures are also caught under natural conditions, such as frontal faces and indoor illumination. The face recognition system of this application can achieve high accuracy without much co-operation from user. The following are the example. Face recognition technology is used to monitor continuously who is in front of a computer terminal. It allows the user to leave the terminal without closing files and logging out. When the user leaves for a predetermined time, a screen saver covers up the work and disables the mouse & keyboard. When the user comes back and is recognized, the screen saver clears and the previous session appears as it was left. Any other user who tries to logon without authorization is denied.

Security: Today more than ever, security is a primary concern at airports and for airline staff office and passengers. Airport protection systems that use face recognition technology have been implemented at many airports around the world.

Image database investigations: Searching image databases of licensed drivers, benefit recipients, missing children, immigrants and police bookings.

General identity verification: Electoral registration, banking, electronic commerce, identifying newborns, national IDs, passports, employee IDs.

Surveillance: Like security applications in public places, surveillance by face recognition systems has a low user satisfaction level, if not lower. Free lighting conditions, face orientations and other divisors all make the deployment of face recognition systems for large scale surveillance a challenging task.

The use of local information partially overcomes some of the typical drawbacks in face recognition (illumination, pose, occlusions), but also opens new questions in turn. Among the critical aspects of local methods, feature selection is the most noteworthy one, as different subjects might have different salient features. Feature selection is of paramount importance as local approaches often generate a feature space with very high dimensionality, which greatly affect the computational complexity of the matching process. Most proposed methods rely on local operators, whose output is generally inputted to a feature selection module or to a dimensionality reduction technique. The way face patterns are represented heavily affects recognition performances, too. Indeed, holistic and local methods generally provide complementary information suggesting that gathering together features produced by both of them should further improve the recognition accuracy.