A critique report on a developed face recognition algorithm 2022

Algorithm development critique
Title
A critique report on a developed face recognition algorithm
Table of Contents
Title.............................................................................................................................................................1
Background..............................................................................................................................................2
Introduction.............................................................................................................................................2
Content....................................................................................................................................................3
Innovation...............................................................................................................................................4
Technical quality......................................................................................................................................5
X factor....................................................................................................................................................5
Presentation............................................................................................................................................7
List of reference.......................................................................................................................................8
1

Algorithm development critique
Background
The paper titled Eigenfaces vs. Fisherfaces: Recognition Using Class Specific Linear Projection
by James and David is chosen for reading since the idea in the paper presented clearly. The
development of an algorithm capable of identifying and authenticating facial images is important
and unique in the science field. The major contribution the paper has made is to develop an
algorithm that is insensitive to light, pose and glass differences and also it has laid a strong
scientific background to the development of other facial recognition algorithm.
Introduction
Technology changes every day. The upgrades in this field of technology are undeniable and
amusing as they consistently produce new innovations and creativity into the field. According to
Chen B 2012 the face recognition system is been used in my many fields of technology
especially in mobile phones industries. Nearly all android phones been produced recently have
the face recognition system install as a way of unlocking the mobile device. According to
Wechsler H 2012 the face recognition knowledge has moved to all dimension of technology and
science fields such as in hospitals, airports and boundary checks, advertisement and also in sports
to verify the identity of the individuals included.
According to Gallagher A.C 2012 the images taken by the face algorithm devices have
improved user interface enabling them to have advanced features including Passive UX (no
activity required), single shot analysis, availability of cross channel inputs, detection of hard
design images (flat images) and also the ability to authenticate natural face activities which are
aimed at improving the securities of the devices having this application. Zhang X and Gao Y
2012 suggests that face recognition is easily accepted by people because it is easy to use and
time saving for instance it just takes a fraction of a minute to get authenticated by the face
recognition algorithms.
Johnston R.A 2009 defined face recognition system as a technical software capable of
identifying a person or a verifying the person’s identity using facial images in order to grant
access to a facility, gadget or to pass a designated security checkpoint. According to Naseem. I,
Togneri. R and Bennamoun. M. 2010 face recognition eigen values matrices can be evaluated by
the use of principal component analysis and the linear discriminant analysis. Various innovations
have been made to improve the ability of these software facial images recognition and
verification in order to increase its security reliance. The innovative attempts includes tasting the
ability of software’s recognition and verification of facial images in different environments
including dark and bright light, detection of like faces and in different weather conditions.
The innovations include the use of both eigenfaces and fisherfaces. Ozen. F 2012 defined
eigenfaces as the values attached to computerized eigen vectors values to represent a person’s
facial image. The facial images are taken such that the probability distribution of the eigen values
are used to form the required covariance matrix by use of principal component analysis. Abidin.
Z. and Harjoko A 2012 defined fisherface as one of the most used algorithms in face recognition
2

Algorithm development critique
which was developed to determine the matrix that increases the ratio between class scatter to the
within class scatter images using elastic map matching on the linear discriminant analysis.
Content
The paper is about development of a face recognition algorithm that is not sensitive to different
variations including pose, glass recognition, light brightness variation and face expression
differences by comparing two methods based on Lambertian surface (the fisherface and
eigenface). Johnson M.K. and Adelson 2009 defined a Lambertian surface as a surface that
reflects light images with the same brightness despite the viewers or the observers viewing angle.
The design used took two approaches using a Lambertian surface. The approaches were:
i. Every image of a Lambertian surface sourced from a constant viewpoint
and illuminated differently lies in a 3D linear subspace. This is because
these Lambertian surfaces have the same light reflectance properties at all
angles.
ii. Face images are not Lambertian materials since they do form their own self
shadows and therefore not all regions of the face will have the same
Lambertian properties.
The paper also gives some insight information about the development of the facial recognition
systems siting that it begun around eighty years ago. R. Fisher designed a face recognition
pattern which uses linear discriminant analysis to verify facial images which was later named
after him to be Fisher linear discriminant analysis. He was motivated by the idea that each
human face has some different and unique facial details, just like in human being recognizing
each other through face looking. Fisher wanted the computer to do the same. His innovation has
today been used and applied almost all fields that uses computer facial vision and facial imagery
identification, also different scholars and scientist have benefitted from his work and developed
other systems related to facial recognition. Eigen face algorithm uses principal component
analysis (mostly used in pattern recognition, where by its objective is to replace the corrected
vectors dimensionally high with the uncorrected vector dimensionally low. It is effective since it
does not use a bigger storage) to compute the face authentication and face identification while
fisher uses linear discriminant analysis to compute the face authentication and face identification.
The major challenge the paper tries to solve is face authentication and face identification.
Authentication is the process of showing that something is true or correct however Fujiwara K.
2011 defined face authentication as the process of granting persons access to something or
somewhere depending on who they are. Guillaumin M. Verbeek and schmid 2009 defined face
identification as the process of recognizing and identifying persons based on facial details. The
paper attempts to test different aspects and variations of facial image recognition in order to
improve face authentication and face identification process. The variations; includes pose,
3

Algorithm development critique
lightning brightness, facial expression and glass recognition tested in comparison of two facial
dimension of face imagery. The two dimensions are eigenfaces and fisherfaces.
Innovation
According to Jain A.K and Li S.K 2011, the face recognition algorithms that were developed
earlier faced many complicated challenges including face authentication and face identification
problems when the set conditions were varied. Changing conditions incudes; light changes,
wearing of glass by the viewer, taking different pose and facial appearance. The algorithms
designed earlier could not perform to the expectations with these adverse conditions. It is to these
challenges that the paper was developed to solve the named challenges. The paper contributes a
facial recognition algorithm.
The author of the paper has created a face recognition algorithm that is not sensitive to light
changes and face appearance using the facial image details. To achieve and realize his target
algorithm, the paper authors performs several experiments using different methods and two
databases namely eigenface and fisherface. The eigenface algorithm takes the following steps:
training set reading, image fitting, selection of the training set, creation of the Matrix by
subtracting the training set face with the mean face image, finding the correlation matrix,
calculating the eigen vectors and eigen faces and finally finding an image with the minimum
Euclidean distance. The eigenface method might have been used in the experiment because it is
independent of the facial geometry and also it’s natural simplicity because it does not need
complex soft-wares for usage.
The first experiment was about the hypothesis testing of an illuminated variable on the face
recognition algorithm that can execute differently and produce good and desired results on a
given Lambertian surface. From the database sourced from the Harvard Robotics Laboratory,
every image used in the experiment was clasped by one of the five subjects (each having sixty
six images) then it was propagated by a designed light source. The light sourced were varied
with an increasing unit angle of fifteen degrees. Ideally the paper aims at reducing the within
class-scatter while preserving the between class-scatter. It was included that with the removal of
the first three PCA (principal components analysis) components of the eigenface resulted in an
improve image quality. Generally considering the experiments done, the fisherface database
method was concluded to be experimentally better than the eigenface database method.
In writing it is unusual for a writer to use a method then to go ahead and site its disadvantages, in
fact most writers ignore the negative parts of the methods they used and only focus on its
advantages. In the paper provided we were able to observe that the writer uses a method called
correlation highlighting that it has the easiest classifier in the nearest neighbor’s formulae. The
writer goes ahead showing how the correlation is applied in imagery: Having an image in the test
set identified by labeling it to the nearest position in the learning set having the length considered
in the image, all the images should have zero mean and unit variance after undergoing a
normalization process. The disadvantages of this correlation method discussed in the paper
4

Algorithm development critique
incudes; that is an expensive process, correlation to be effectively needs large amounts of storage
and the lastly that the process of changing the lightness would be resulting in images that are
combined images and would therefore require a complex the processes in the learning set.
Technical quality
The technical quality in the paper is of high quality. This is due the excellent use of the training
sets and the learning set, use of enough and moderately required experiments to make
conclusions, it has a coherent flow, and for instance the paper could be easily read and
understood by a reader.
According to Domingos P. M 2012, the main idea behind machine learning is to predict more
results beyond the experiments or examples in the training set. In the paper on the eigenface
method correlation it is clearly seen that the correlations of the eigenfaces matches the size of the
training set. As the experiments proceeds, the author does not expect the eigenface correlation to
improve. Also on the fisherface method lightness experiment, the paper shows that while
attempting to minimize the within class-scatter for every available class, then it was expected
that it can relay projections directions that can identify facial images that are superior to those in
the training set.
In the paper three experiments were done testing different aspects namely: changes in light,
varying of facial expression and eye wear and testing the recognition of the glass. The way and
the depth at which each experiment was carried out were convincing. For example the
experiment testing the lights variation was done by sourcing data from Harvard Robotics
laboratory. A total of three hundred and thirty images were used, from five persons, each
contributing sixty six images. The images were then propagated by a light bulb, altering the
angles by increasing it by fifteen degrees. Also in the facial images recognition experiment the
number of images were sourced from sixteen persons with varied gender, age and facial
condition including presence of hair and glass conditions. Considering how these experiments
were done it is sufficient to say that the results obtained were significantly accurate thereby
making the conclusions raised also accurate and efficient.
The technical quality of the paper is visible on how the work is presented. The paper work can be
understood easily and the basic idea of the paper be seen at glance with the understanding of how
the problem of face authentication and face identification it tries to solve by comparing the two
methods; eigenface and fisherface. This is further supported with the authors cited in the paper
including Hallinan, Moses et al, Nayar and Murase who contributed differently to the writing of
the paper.
X factor
The development of a face recognition algorithm using specified class linear projections is
appropriate using the eigenface and the fisherface methods. It is appropriate because
comprehensive experiments encompassing the usage of over three hundred images were done on
5

Algorithm development critique
a complete comparison of the two techniques to give significant conclusions that led to the
development of the new face recognition algorithm.
Development of such great application such as face recognition algorithm that can identify and
authenticate faces can be applied in many aspects of the computer vision. According to
Hassaballah M and Aly S 2015 face recognition could be used in advertising to identify potential
markets like a detected face of a young girl would be shown make-ups ads while the face of a
young boy would be shown ads relating to games and movies. Face recognition can also be
applied in security systems whereby the systems checks would be fitted with face scanners that
would only allow access to permitted individuals, this algorithm can be used in payments and
checks clearances to facilitate faster means of payments as opposed to the old signing style. Jafri
R and Arabnia 2009 suggested that face recognition in the healthcare departments can be an
important utility that can be used in determining patient’s age, sickness and other bio-data
required. This is important as it saves the doctors time and efforts. In African countries many old
people do not know their dates of birth, thereby just guessing their ages. This face recognition
can help doctors know the exact age of such individuals. Parmar D.N 2014 suggested that the
airports and boundary checks should apply the face recognition technologies in order to boost
security through severe identification and authentication of persons passing the security check
points.
According to Ramchandra A and Kumar R 2013 and Bhatia R 2013 the use of facial recognition
systems as a security utility is still not safe enough since it can fail in extreme conditions such as
identical twins face might look exceedingly similar for the scanners, also lightness problems and
different poses. To boost services rendered by face recognition system, Parkhi O.M 2015
recommended the importance of supporting face identification and verification processes by
other biometric appliances such as the fingerprint scanner which is more secure compared to the
face recognition. Masuda M 2012 recommended that face recognition system should have its
scanners position at an angle than would not be affected by lights or should have apparatus that
controls the light and taking care of the different poses that might be observed.
The experiments done in the paper could spark a good discussion in class. The students or
readers can discuss, demonstrate, and redo the experiments just to confirm the results. What’s
interesting about the paper is the innovation of the new algorithm using by comparing two
methods, also how fisher develop the fisherface method is also interesting. The students can be
asked to suggest ways to improve the developed algorithm or develop a new face recognition
algorithm using different methods.
6

Algorithm development critique
Presentation
The presentation in the paper was fairly done, the structure is clear. The arguments, statements and
discussions in the text were not easily understood. Some points were not clearly outlined for instance in
the explanation of the developed of the face recognition algorithm, much emphasis was given to the
comparison of the two methods used fisherface and eigenface using the experiments done. Very little
discussion was given on the algorithm developed and its development.
For the paper to have a good presentation the report should contain the following; a good research
design which is instrumental in answering the research questions, a good research criteria that
stipulates how the experiments in the research should be carried out, a clearly designed research
methods having all the data of the participants data analysis and experimentation formulas and should
also contain a discussion and conclusion part. The paper studied did have the parts mentioned but were
not fully implemented as the discussion and conclusion parts were incomplete.
The presentation in the paper could be improved by having the ideas and the thoughts of the author
written in an easy and coherent form. Also it can be made better by giving enough explanation to points
and discussion.
7

Algorithm development critique
List of reference
Abidin, Z., & Harjoko, A. (2012). A neural network based facial expression recognition
using fisherface. International Journal of Computer Applications, 59(3).
Bhatia, R., 2013. Biometrics and face recognition techniques. International Journal of Advanced
Research in Computer Science and Software Engineering, 3(5).
Chen, B., Shen, J. and Sun, H., 2012, May. A fast face recognition system on mobile phone.
In 2012 International Conference on Systems and Informatics (ICSAI2012) (pp. 1783-1786).
IEEE.
Domingos, P. M. (2012). A few useful things to know about machine learning. Commun.
acm, 55(10), 78-87.
Fujiwara, K. (2011). U.S. Patent No. 7,974,446. Washington, DC: U.S. Patent and Trademark
Office.
Gallagher, A.C., Loui, A.C., Cerosaletti, C.D., Hibino, S.L., Das, M. and Stubler, P.O., Eastman
Kodak Co, 2012. User interface for face recognition. U.S. Patent 8,315,463.
Guillaumin, M., Verbeek, J., & Schmid, C. (2009, September). Is that you? Metric learning
approaches for face identification. In 2009 IEEE 12th international conference on computer
vision (pp. 498-505). IEEE.
Hassaballah, M. and Aly, S., 2015. Face recognition: challenges, achievements and future
directions. IET Computer Vision, 9(4), pp.614-626.
Jafri, R. and Arabnia, H.R., 2009. A survey of face recognition techniques. Jips, 5(2), pp.41-68.
Jain, A.K. and Li, S.Z., 2011. Handbook of face recognition. New York: springer.
Johnson, M.K. and Adelson, E.H., 2009, June. Retrographic sensing for the measurement of surface
texture and shape. In 2009 IEEE Conference on Computer Vision and Pattern Recognition (pp.
1070-1077). IEEE
Johnston, R.A. and Edmonds, A.J., 2009. Familiar and unfamiliar face recognition: A
review. Memory, 17(5), pp.577-596.
8

Algorithm development critique
Masuda, M. and Nagao, K., Oki Electric Industry Co Ltd, 2012. Face recognition system. U.S.
Patent 8,340,366.
Naseem, I., Togneri, R., & Bennamoun, M. (2010). Linear regression for face recognition. IEEE
transactions on pattern analysis and machine intelligence, 32(11), 2106-2112.
Özen, F. (2012). A face recognition system based on eigenfaces method. Procedia Technology, 1,
118-123.
Parkhi, O.M., Vedaldi, A. and Zisserman, A., 2015, September. Deep face recognition.
In bmvc (Vol. 1, No. 3, p. 6)..
Parmar, D.N. and Mehta, B.B., 2014. Face recognition methods & applications. arXiv preprint
arXiv:1403.0485.
Ramchandra, A. and Kumar, R., 2013. Overview of face recognition system
challenges. International Journal of Scientific & Technology Research, 2(8), pp.234-236
Wechsler, H., Phillips, J.P., Bruce, V., Soulié, F.F. and Huang, T.S. eds., 2012. Face recognition:
From theory to applications(Vol. 163). Springer Science & Business Media.
Zhang, X. and Gao, Y., 2009. Face recognition across pose: A review. Pattern recognition, 42(11),
pp.2876-2896.
9