See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/318591109

Automatic Face Naming Using Image Processing: A Review Study Based on

Face Name Detection in Images or Videos

Conference Paper  in  Communications in Computer and Information Science · July 2017

DOI: 10.1007/978-981-10-5427-3_38

CITATIONS READS

0 225

3 authors:

Pragya Baluni Shashi Kant Verma

thdc ihet Govind Ballabh Pant Engineering College
6 PUBLICATIONS   13 CITATIONS    62 PUBLICATIONS   360 CITATIONS   

SEE PROFILE SEE PROFILE

Yp Raiwani
Hemwati Nandan Bahuguna Garhwal University
19 PUBLICATIONS   72 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Cloud Service scheduling in VANET View project

Data mining Research View project

All content following this page was uploaded by Pragya Baluni on 01 March 2019.

The user has requested enhancement of the downloaded file.

 Automatic Face Naming Using Image
Processing: A Review Study Based on Face
Name Detection in Images or Videos

Pragya Baluni1(&), S.K. Verma2, and Y.P. Raiwani3

1
Department of Computer Science and Engineering, THDC IHET,
Tehri Garhwal, India
pragyabaluni@ymail.com
2
Department of Computer Science and Engineering, GBPEC, Pauri, India
skverma.gbpec@rediffmail.com
3
Department of Computer Science and Engineering, HNBGU, Srinagar, India
yp_raiwani@yahoo.com

Abstract. Automatic character identification in images is essential for semantic

image analysis such as indexing, summarization and retrieval. Face identifica-
tion, however greatly characteristic to individuals, is an enormously troublesome
errand in modernized vision. Structure is enthused about modified naming of
people in different images or montage material with their character. A note-
worthy testing issue due to the extensive assortment in intrapersonal appearance
assortments in light of the fact that photometric variables, for instance, dis-
tinctive lighting tints, viewpoint point, and scale, or due to changes in expres-
sions, or impediments, hairstyles, so appearances of each characters and the
inadequacy vulnerability of available comment. The objective of this paper is to
clarify the characteristics required in programmed face naming procedure and to
review some of their recent studies.

Keywords: Automatic face naming
Face identification
Character

recognition
Review
Feature extraction

1 Introduction

Recent studies on the area, propose not to utilize just the graphic or literary substance
alone, yet to consolidate them both. With this methodology, face acknowledgment
issue is disentangled to a face-name affiliation issue. Taking after these methodologies,
in our strategy literary and visual data is joined to name faces. A directed algorithm is
utilized for naming a predetermined number of classes having a place with all the more
often showing up appearances. We partition the issue into 2 sub issues; one the more
habitually seeming confronts the less-much of the time showing up countenances on
the web images are named. Another the countenances that are not coordinated with any
classification are then thought to be the less-as often as possible showing up appear-
ances and named utilizing the literary substance. We extricated every one of the names
from printed substance, and after that wipe out the ones used to name as often as

© Springer Nature Singapore Pte Ltd. 2017

M. Singh et al. (Eds.): ICACDS 2016, CCIS 721, pp. 357–368, 2017.
DOI: 10.1007/978-981-10-5427-3_38
358 P. Baluni et al.

possible showing up countenances some time recently. The rest of the names are the
competitor classes for less as often as possible showing up countenances.
Face identification is a procedure that decides the area of a human face in a
computerized image. In this venture, face recognition is the initial step. It includes
catching images continuously from a camera and after that figuring out if or not the
image contains facial elements. On the off chance that a face is distinguished in an
image it is then highlighted in the image and went on to face acknowledgment.
Fingerprints, hand geometry, iris checks, DNA examination and to some degree
individual marks are all biometric identifiers. In any case, the stand out that does not
defer or meddle with access is face acknowledgment. People distinguish others by their
face and voice and in this manner are liable to be more alright with a framework that
utilizations face and voice acknowledgment [2].
This makes face acknowledgment perfect for high activity regions which are
interested in the overall population for e.g. airplane terminals and railroad stations,
ATM’s, open transportation and organizations of numerous types. Face acknowledg-
ment gives a record of who was there. Since the record is stored in a database, known
persons can be detected automatically and unknown persons checked quickly.
Face tracking and face clustering procedure as shown in Fig. 1 might fizzle because
of impediment, low determination video, substantial movement, changed foundation
and different conditions. Face identification of image is more simple than video. The
circumstance is more terrible in video. This conveys commotions to the character
identification in video. Additionally different performing artists assuming part for
various periods of the same character name. There might be different represents, a ton
of expression and light impact, dressing, wearing, even roll out up and hairstyle
changes [3]. The significant intra-class difference, the similar individual name relates to
the characteristics of massive variation presences. The assurance for the quantity of
indistinguishable appearances is not trivial.

Fig. 1. Example of face recognition [1]

1.1 General Attributes for Face Naming

Choosing face tracker can be a troublesome assignment due to the accessibility of
numerous face trackers. The application supplier must choose which face tracker is
 Automatic Face Naming Using Image Processing 359

most appropriate to his/her individual needs and, obviously, the kind of video that
he/she needs to use as the objective. As a rule, the critical issues are the tracker’s
velocity, vigor, and precision. Like the instance of numerous preparing apparatuses for
communicate video, velocity is not the most basic issue on the grounds that discon-
nected handling is allowed in most video organizing and ordering exercises. Be that as
it may, an ongoing face tracker is vital if the objective document depends on too huge
an amount of video, say, 24 h of consistent video recording that requirement day by
day organizing [15]. Additionally, the pace of the tracker is basic in most
non-communicate video applications, e.g. HCI. There is dependably a tradeoff amongst
velocity and execution related issues, for example, robustness and accuracy.

1.2 Image-Based Features

The aforementioned degradations manifested in biometric tests can be surveyed uti-
lizing image highlights that are computationally inexpensive to compute. Auto-
matic QA is basically tended to by breaking down spatial and temporal components
that are demonstrative of the image content [16]. Highlights that are utilized widely as a
part of current writing can be extensively isolated into four classifications:
• Orientation elements are acquired from edges in the image. If there should be an
occurrence of the iris and face, edge data is generally utilized as elements for
acknowledgment. Obscuring, brightening, and commotion corrupt edge data along
these lines influence execution. Subsequently, introduction data can give a decent
sign of the nature of a biometric sample.
• Power range is a transient measure of the force of the image signal. This measure
means that the measure of data present in image area. Thus, otherworldly vitality is
frequently registered for various image areas to acquire nearby evaluation of value.
• Intensity statistics are immediate measurable assessment of powers of pixels in the
image. Normally, a factual measure, for example, Kurtosis or point spread function
(PSF) estimation is utilized to evaluate obscuring or enlightenment corruption in the
image. The measure can then be contrasted with the reference values got from
perfect images to figure the degree of degradation.
• Wavelet transform gives both spatial and recurrence comprehension of the data
content in every sub-band of the image. These are especially suited to learn the
nearness of fine miniaturized scale edges in the iris area and to acquire nearby
examination of value in various areas of image.

1.3 Facial Features

It’s hard to accomplish a good precision rate of naming appearances for images caught
under abandoned situation. Not just brightening, posture, center resolution of images
source to bad naming in face acknowledgement systems, but too eye-classes, make-up,
facial hairs, hairstyle a few other individual and atmosphere associated changes on the
face, create acknowledgment troublesome for analysts of the range. With a specific end
360 P. Baluni et al.

goal to lessen the impacts of such issues, highlight vector choices on facial depictions
are essential. Two unique techniques for face representations are connected for this
study. A novel methodology presented by Kumar et al. [13], and the another is the
SIFT [17] descriptors extricated for 9 particular facial focuses, as its label shows a
measure- invariant technique vigorous to stance, enlightenment, measure and so forth.

2 Background of the Study

As such, it is a significant requirement for mechanical accomplishments particularly in

business and law implementation applications. Individuals have a tendency to enhance
techniques for applications that need security and protection as opposed to utilizing
passwords. Though diverse types of particular biometric identification are present, for
example, iris or retinal or fingerprints, are expressed in the investigation of Duygulu
et al. [4], these techniques depend on the support of the issue of individual. In this way,
with the growing interest on innovative change, face acknowledgment is about to
become an integral part of our lives.
The prominent individuals, for example, big names or government officials seem all
the more often on the internet, consequently; gathering image testers for them is a
moderately less demanding errand contrasted with the assignment of gathering image
tests for individuals seem less-as often as possible on the web. With respect to cer-
tainties, in this study we concentrate on naming less every now and again showing up
individuals who are seen alongside all the more as often as possible showing up
individuals on the web. The procedure will be just to name the all the more as often as
possible showing up individuals to start with, then name the less-regularly seeming
ones by them, utilizing the textual content [5–7].
Ozkan and Duygulu in [8] relate a graphics based technique for naming appear-
ances. Dissimilar to present face depictions with concern focuses, not just specific
focuses on the face are utilized, additionally identified interest focuses are considered.
Then again, the utilization of all recognized interest focuses on a face, instead quite
compelling focuses; don’t generally give the best result. At long last, utilizing their
closeness diagram, the most comparative images are discovered by means of delivering
the deepest subsection on that chart that drives the insignificant images to be removed.
Guillaumin et al. in [9] proposed a strategy of naming the faces from their database
that comprises of newscast photographs with inscriptions by considering two situa-
tions, one is to discover faces for a solo query. The main methodology is a kNN
technique with an edge and the additional method is to separate among the neighbors.
With the utilization of nine specific facial elements, they beat the coordinating attention
point’s issue experienced in [10].
Le and Satoh in [11] present an unverified technique for explanations of counte-
nances on the internet. Their technique comprises of 2 stages. The first stage is to
source the information from the web and locate the deepest set by means of positioning
the conveyance of visual likenesses and the 2nd stage is the grouping of the output
query where coupled naming of appearances as craved individual or non-sought
individual is resolved. This strategy is enhanced utilizing SVM alleged Lib-SVM [12].
 Automatic Face Naming Using Image Processing 361

Kumar et al., in [13], propose a technique for relationship of names to confront

utilizing better reasonable dataset that is not the same as current face acknowledgment
dataset in the feeling of countenances being caught “in nature”. Their dataset comprises
of 30.281 images identified from half a million captioned images.
Le et al. in [14] additionally centered on name-face affiliation; though as opposed to
the overall methodology of doling out names to the countenances on the images, they
intend a strategy which plans to accomplish a coordinated task for names and faces.
They concoct this thought taking into account the case that not every one of the names
extricated from literatures, or not every one of the countenances separated from images,
have the same probability for being critical as far as having a relating face-image in
light of the fact that the name is said, imagedness, or having a comparing name in light
of the fact that the image contains the face, namedness. The tabular analysis of different
face acknowledgment studies is appeared underneath in Table 1.

Table 1. Tabular analysis of different face acknowledgment studies.

Author Technique name Methodology used Results
name/year
Duygulu Object recognition as Diverse types of particular To enhance techniques for
et al. (2002) machine learning biometric identification are applications that need security
used, for example, iris or retinal and protection as opposed to
or fingerprints utilizing passwords
Ozkan and Graphics based technique Utilization of all recognized The most comparative images are
Duygulu interest focuses on a face, discovered by means of
(2006) instead quite compelling delivering the deepest subsection
focuses on that chart that drives the
insignificant images to be
removed
Guillaumin Automatic face naming with Proposed a strategy of naming With the utilization of nine
et al. (2008) caption-based supervision the faces from their database specific facial elements, they beat
that comprises of newscast the coordinating attention point’s
photographs with inscription issue
Le and Unsupervised face Proposed strategy is enhanced Present an unverified technique
Satoh annotation by mining the utilizing SVM alleged for explanations of countenances
(2008) web Lib-SVM on the internet
Kumar Attribute and simile Propose a technique for Dataset comprises of 30.281
et al. (2009) classifiers for face relationship of names to images identified from half a
verification confront utilizing better million captioned images
reasonable dataset
Le et al. Finding important people in A strategy which plans to Centered on name-face
(2007) large news video databases accomplish a coordinated task affiliation; though as opposed to
using multimodal and for names and faces the overall methodology of
clustering analysis doling out names to the
countenances on the images

3 Methodologies
3.1 Phase-Only Correlation (POC)
Phase-Only Correlation (POC) has been likewise called stage relationship, which has
one of the image coordinating methods and has been effectively actualized to biometric
362 P. Baluni et al.

confirmation and PC vision challenges. The POC capacity has been characterized as
reverse Discrete Fourier change of the standardized cross force range. The tallness and
location of the relationship top has been shown the likeness and translational devel-
opment between images, individually. The algorithm utilizing POC has been proposed
for the face acknowledgment [18]. The POC-based face acknowledgment algorithm has
been taken after exceptionally basic methodology. To begin with, the face image has
been standardized and additionally the rest venture of the LBP-based face acknowl-
edgment algorithm. Following, an arrangement of the position focuses has been situ-
ated on the face image to assess the neighborhood piece similitude. The stage based
correspondence coordinating has been utilized; the comparing point sets are acquired,
where the relating point pair having low closeness esteem has been disposed of as an
anomaly. At last, the coordinating score has been computed as the quantity of right
relating point sets. In the event that the image has blocked district, the coordinating
score of the POC coordinating has been diminished notwithstanding for the certified
pair [19].

3.2 Local Binary Pattern (LBP)

Basically, the local binary pattern (LBP) had been intended for surface portrayal. It has
expanded increasingly researchers “and faculties” consideration for its computational
adequacy and high separation property additionally it has been invariant to monotonic
dark scale changes which are essential for surface investigation. Other than that, the
likelihood of preparing image continuously has been empowered by its computational
straightforwardness which beats numerous other surface descriptors [20]. LBP is a
standout amongst the best nearby component extractors, which removes surface ele-
ments of the photo by contrasting every pixel and nearest in a little neighborhood.
Besides, there is no preparation necessity, which makes it quick and simple to incor-
porate into the new information sets. Besides, in light of the utilization of histograms as
the capabilities, it has vigorous against revolution and scaling [18].
Additionally, the sample-image size measurement can be decreased to the quantity
of histogram canisters. As the dim estimations of neighbors have been thought about, it
is strong beside monotonic changes in the face image. Accordingly, LBP is one of the
splendor obtuse descriptors. The strategy of the LBP based face acknowledgment
algorithm is exceptionally straightforward. Initially, the face district has been recog-
nized, the component focuses on the face have identified, and after that the scale,
revolution and interpretation have standardized by position of highlight focuses. Next,
LBP has been computed for every nearby piece image. At last, the coordinating score
has been computed utilizing the weighted Chi square separation between histograms of
LBPs [21] Different strides of LBP
i. Radius of 2 neighboring windows and total count of neighbors.
ii. Binary designs acquired by correlation among the middle pixel and its neighbors.
iii. Utilizing parallel ciphers to create the histograms and link them for every image
sub-piece LPQ [20].
 Automatic Face Naming Using Image Processing 363

3.3 Local Phase Quantization (LPQ)

LPQ is a well-known histogram-based component extractor which has a place with the
group of nearby surface descriptors, and played out the evaluation of stage in a
neighborhood window at the pixel position. In recurrence space, Local stage investi-
gation has been directed to a point by point shine coldhearted surface depiction of face
images. LPQ has been additionally uncaring against image debasement, obscure
impact, which happens for the most part in true applications, for example, video
reconnaissance, which has been created by out of center of camera or item movement.
LPQ has a typical splendor and obscure heartless component extractor. To utilize
nearby stage quantization for face portrayal, it can be connected for the methodology of
face acknowledgment [21]. To begin with the face image has been named with the LPQ
administrator. At that point, the name image has been separated into non-covering
rectangular areas of equivalent size and a histogram of marks has been figured
autonomously inside every district. At last, the histograms from various districts have
been connected to fabricate a worldwide depiction of the face. LPQ strategy has been
given below [20]
i. STFT at four particular frequencies incorporated a little space at the pixel location.
ii. Based on the indication of genuine and nonexistent qualities, Binary quantization
has been finished.
iii. A histogram has been directed taking into account the 8-bit stream.
iv. Finding the genuine and nonexistent coefficients for the four frequencies.

3.4 Local Binary Patterns & Nearest Feature Space (LBP&NFS)

The simple and efficient Nearest Feature Space (NFS) and Local Binary Patterns
(LBP) technique classifiers that the great speculation capacity has joined to build
another methodology for fulfilling the ongoing and power choice execution. The face
image database got from web has been utilized for analyses. The handy results
demonstrate that the model technique performs superior to anything conventional
approaches as far as continuous productivity (0.7 s for each sample) and precision
(72.5% Average Recognition Rate) NFS was initially shown by Jen Tzun and actu-
alized in face acknowledgment. Basically, it has been an augmentation of nearest
neighbor (NN) in light of the fact that it has been suited a bigger ability of model
components by amplifying the geometrical parts of indicate point and space. Essen-
tially, it has been tackle with separation between testing highlight point and its forecast
highlight point in extended component space as per the geometrical ideas. As NFS
demonstrations need than closest component plane (NFP) and closest element line
(NFL) by Chien presumed that it has been just centered on the NFS-related strategies.
The work has been utilized Local Binary Pattern (LBP) and NFS which has been an
augmentation of multi choice level blend in light of consolidated elements for face
acknowledgment procedures.
For pre-processing of images, the histogram equalization (HE) strategy has been
utilized to dispose of enlightenment impact since utilizing the current ways to deal with
364 P. Baluni et al.

face acknowledgment. The light standardization has been performed independent of the
lighting up conditions then the two components are joined, the LBP has been utilized to
remove further element on DCT-Scharr-image. At that point Nearest Neighborhood
(NN) classifier has been to discovering N quantities of the ideal applicants. The
course-to acceptable structure has been separated into the rough and the fme stages. In
this way, before the termination of this stage the rough stride can wipe out a ton of
no-great examples in the Training Databases and diminishing the extension to character
the test image of the N ideal applicants. At long last, the utilization of all the more
capable progressed LBP again and the extra powerful speculation capacity of Nearest
Feature Space (NFS) to get the personality of the test image. The NFS not just broaden
the model of NN, and increment the limit of speculation additionally the reasonable set
number of applicants circumstance is ideal for occupied in concert the execution of the
NFS. The NN expansion of the NFS broadens the model limit from looking the closest
component opinion to the closest separation amongst question and model space.
Indisputably, addition to this part choice level combination part, we join both
component level combination and choice level combination successively to frame the
new compelling and continuous face acknowledgment strategy. Exploratory conse-
quences with average response rate (ARR) of seventy six percentage and 0.7 s for
every image of constant productivity output in light of the Luminary Database [22].

3.5 Local Binary Pattern & Local Phase Quantization (LBP&LPQ)

A local-based brightness insensitive algorithm has been generally utilized for face
acknowledgment which has the course of action of image standardization and invariant
descriptors. The Lighting unfeeling show of picture has been acquired in view of the
proportion of inclination sufficiency to the first image force and isolated into littler
sub-squares. Neighborhood stage quantization and Multi-scale nearby paired example,
separate the sub-pieces qualities. Separation of nearby closest neighbor classifiers has
been measured and melded at the score level to locate the finest match and choice level
combination consolidates the aftereffects of two coordinating systems joining highlight
extractors can bring about preferable execution over utilizing one descriptor. Amongst
the nearby descriptors with huge execution, LPQ and LBP have recognized not
computationally quickly with basic estimations which are inhumane beside monotonic
image corruption [20]. The examination of different face acknowledgment algorithms
on the premise of their exactness and diverse arrangement of information base is
appeared underneath in Table 2.

Table 2. Comparison of algorithms using different database based on accuracy.

Algorithm Database Accuracy (%)
POC FERET 96%
LBP YALE 96.97%
LPQ YALE 95.3%
LBP & NFS ORL 72.5%
LBP & LPQ YALE 98.3%
 Automatic Face Naming Using Image Processing 365

4 Review Summary

There are some conclusions that could be drawn after reviewing the different face
identification methods of recent research developments that are as follows:
State-of-the-Art
The effective techniques are the model-based which coordinate point of interest proofs
from nearby fixes with a worldwide shape requirement. The coordination of the
neighborhood and worldwide data or Ist-and 2nd-level processes is acknowledged with
differing qualities of techniques extending from Bayesian forecast to SVR. The
two-level methodologies are techniques that in the primary level concentrate fiducially
historic points, and in the second level anticipate and merge milestones with less
enlightening elements under direction of a face shape model are more fruitful. It gives
the idea that the execution of algorithms in the most recent five years have enhanced to a
point where for the m 17, it is on a standard with manual area stamping. Truth be told, on
the off chance that we constrain our perceptions to the distributed results in the articles
this implies a couple rate focuses, as 2–3% of IOD. External the m 17 set, the exactness
stays inside 5–8%. In any case, our tests on the seven most conspicuous area detecting
algorithms have uncovered that these outcomes are not generally reproducible.
Identification Under Adverse and Realistic Conditions
Face identification methods, being regularly neighborhood in nature, can be made
heartier to characteristic inconsistency and securing circumstances. It is conceivable to
express that light impacts can be for the most part remunerated by such preprocessing
ventures as Laplacian of Gaussian (LoG) separating or histogram balance. Addition-
ally, outward appearances and unobtrusive stance varieties can be compensated for by a
wealthier arrangement of preparing occurrences. The most despicable aspect of area
stamping stays serious stance varieties, i.e., past 20° yaw edges and tilts, particularly
when self-impediments happen. We expect that in-plane revolutions can be revised
after discovery of the face and of the eyes. It creates the impression that half and half
techniques like appearance-helped geometry-based strategies [21, 22], 3D-helped
strategies [23, 24] or an associated battery of neighborhood layouts as in [25] hold a
decent guarantee for achievement.
Ground-Truthed, Robust and Databases
The positive outcomes of various algorithms may vary emphatically from one database
to another database. Actually, the over database execution of the early approaches, as
they were ready on one database and then afterward tried on another database,
demonstrated a shortcoming i.e. causing severe descents in execution. It is urging to
observe those later approaches, quite [24, 26] have vigorous execution over various
distinctive databases. The test comes about, however very broad in aim, has not yet
uncovered a definitive and most reasonable correlation, in that techniques have not
been offered opportunities to be prepared on self-assertive blends of databases. Truth be
told, for a reasonable correlation, we recommend that strategies ought to be tried in the
Leave One Database Out style (LODBO), where algorithms are prepared with different
databases aside from one and after that tried on the prohibited one. To conclude, this
review of procedural examinations and the point of interest databases ought to be
366 P. Baluni et al.

stretched out to element acts to assess the attendant issue of historic point following
algorithms [27]. Truth be told, the historic point following issue itself justify a distinct
review work.
Methods to be Reconnoitered
It could be promising examination ways in area stamping strategies are the accom-
panying: (i) Sparse word references: The worldview of acknowledgment beneath
sparsity limitation and working of prejudicial lexicons appears to be one reasonable
technique. The discriminative scanty word reference can be developed per milestone
[28, 29] or by and large as in [30]; (ii) Adaboost chose highlights for multi-view
land-stamping: Gabor or Haar wavelet highlights chose through adjusted Adaboost
plan where shared characteristic and geometric setup of historic point appearances is
misused [31]; (iii) Multi-outline land-detecting: Determination of point of interest
positions abuses the data in ensuing casings of a video, utilizing, for instance, a
spatio-transient depictions [32, 33].
Data Mining in Related to Gesture and Facial Expression
At present web contains no less than two lakhs face recordings [34], as a rule clarified
with logical data, and this number is quickly expanding. This abundance of information
gives an intriguing chance to investigate human outward appearances; it could be said,
to information mine demeanors crosswise over societies, sexual orientations, features
and ages. This wellspring of face information is essential since it has to know as
attention to that absence of naturalistic, unconstrained demeanor information was a
noteworthy barricade in PC examination of outward appearances. It has been called
attention to those pretending expressions that is outward appearances carried on as
incited by a controller contrast in their progression and assortment when contrasted
with unconstrained articulation of the same feelings. We accept strong area detecting
will be involved for appointing this exceptionally rich web wellspring of honest to
goodness hominoid expressions [35].

5 Conclusion

This paper gives an itemized depiction of different face expectation models by utilizing
image preparing from which specialist can get a thought for a productive face forecast
procedures considering maturing consequences for human appearances. This paper has
examination made between existing models on the premise of exactness, velocity, utilized
strategies, and database size. There are prospects extents of enhancements in present
approaches as no model certification hundred percent precision and are limited to particular
element. Present philosophies can be extended by joining face forecast and acknowl-
edgment, so that a hearty acknowledgment based expectation model can be created
keeping in domain the impacts of maturing, facial qualities, environment and so on that is
fit for perceiving current info image that is not present in the image database by anticipating
it on the premise of already put away images that comparing to current information image.
Taking everything into account, facial area detecting has progressed significantly
from its small starting toward the end of 80. The issue can be thought to be understood
for close forward countenances with impartial to gentle facial signs, and sufficient
 Automatic Face Naming Using Image Processing 367

determination. It creates the impression that a portion of the effective approaches can be
keep running at images rates at moving conditions. Then again, for uncontrolled
conditions including subjective postures and expressions, the issue can’t yet be con-
sidered as completely tackled. Late research comes about, be that as it may, however,
give us a positive outlook.

References
1. Berg, T., Berg, A., Edwards, J., Maire, M., White, R., Teh, Y., Learned Miller, E., Forsyth,
D.: Names and faces. Technical report, University of California Berkeley (2007)
2. Bicego, M., Lagorio, A., Grosso, E., Tistarelli, M.: On the use of sift features for face
authentication. In: CVPRW 2006: Proceedings of 2006 Conference on Computer Vision and
Pattern Recognition Workshop, p. 35. IEEE Computer Society, Washington, DC (2006)
3. Chang, C.-C., Lin, C.-J.: LIBSVM: a library for support vector machines (2001). Software:
http://www.csie.ntu.edu.tw/cjlin/libsvm
4. Duygulu, P., Barnard, K., de Freitas, J.F.G., Forsyth, D.A.: Object recognition as machine
translation: learning a lexicon for a fixed image vocabulary. In: Heyden, A., Sparr, G.,
Nielsen, M., Johansen, P. (eds.) ECCV 2002. LNCS, vol. 2353, pp. 97–112. Springer,
Heidelberg (2002). doi:10.1007/3-540-47979-1_7
5. Everingham, M., Sivic, J., Zisserman, A.: Taking the bite out of automatic naming of
characters in TV video. Image Vis. Comput. 27(5), 545–559 (2009)
6. Huang, G.B., Ramesh, M., Berg, T., Learned-Miller, E.: Labeled faces in the wild: a
database for studying face recognition in unconstrained environments. Technical report
07-49, University of Massachusetts, Amherst (2007)
7. Friedman, J.H.: Another approach to polychotomous classification (2006)
8. Ozkan, D., Duygulu, P.: A graph based approach for naming faces in news photos, vol. II,
pp. 1477–1482 (2006)
9. Guillaumin, M., Mensink, T., Verbeek, J., Schmid, C.: Automatic face naming with
caption-based supervision, pp. 1–8 (2008)
10. İkizler, N., Duygulu, P.: Person search made easy. In: Leow, W.-K., Lew, Michael S., Chua,
T.-S., Ma, W.-Y., Chaisorn, L., Bakker, Erwin M. (eds.) CIVR 2005. LNCS, vol. 3568,
pp. 578–588. Springer, Heidelberg (2005). doi:10.1007/11526346_61
11. Le, D.-D., Satoh, S.: Unsupervised face annotation by mining the web. In: ICDM, pp. 383–
392 (2008)
12. Kre, U.H.-G.: Pairwise classification and support vector machines. pp. 255–268 (1999)
13. Kumar, N., Berg, A.C., Belhumeur, P.N., Nayar, S.K.: Attribute and simile classifiers for
face verification. In: IEEE International Conference on Computer Vision (ICCV) (2009)
14. Le, D.-D., Satoh, S., Houle, M.E., Nguyen, D.P.T.: Finding important people in large news
video databases using multimodal and clustering analysis. In: ICDEW 2007: Proceedings of
2007 IEEE 23rd International Conference on Data Engineering Workshop, pp. 127–136.
IEEE Computer Society, Washington, DC (2007)
15. Miller, T., Berg, A.C., Edwards, J., Maire, M., White, R., Teh, Y.-W., Learned-Miller, E.,
Forsyth, D.: Faces and names in the news. In: IEEE Conference on Computer Vision and
Pattern Recognition (CVPR) (2004)
16. Lu, K., Dong, L.: Using LBP histogram for face recognition on Android platform, pp. 266–
268. IEEE (2011)
368 P. Baluni et al.

17. Tajima, Y., Ito, K., Aoki, T.: Performance improvement of face recognition algorithms using
occluded-region detection. IEEE (2013)
18. Nikan, S., Ahmadi, M.: Local gradient-based illumination invariant face recognition using
local phase quantization and multi-resolution local binary pattern fusion. Inst. Eng. Technol.
(IET) Image Process. 9(1), 12–21 (2014)
19. Ahonen, T., Hadid, A., Pietikäinen, M.: Face description with local binary patterns:
application to face recognition. IEEE T-PAMI 28(12), 2037–2041 (2006)
20. Haixu, W.L., Li, C.X.: A new face recognition algorithm using LBP and NFS. In:
International Conference on Computer Design and Applications (ICCDA), vol. 1, no. 6,
pp. 323–327. IEEE (2010)
21. Wiskott, L., Fellous, J.M., Kruger, N., von der Malsburg, C.: Face recognition by elastic
bunch graph. IEEE Trans. Pattern Anal. Mach. Intell. 7, 775–779 (1997)
22. Salah, A.A., Çınar, H., Akarun, L., Sankur, B.: Robust facial land marking for registration.
Ann. Telecommun. 62(1–2), 83–108 (2006)
23. Lu, X., Jain, A.K.: Automatic feature extraction for multi-view 3D faces recognition. In:
Proceedings of International Conference on Automatic Face and Gesture Recognition,
Southsampson, UK (2006)
24. Dibeklioǧlu, H., Salah, A.A., Akarun, L.: 3D facial landmarking under expression, pose and
occlusion variations. In: Proceedings of IEEE International Conference on Biometrics:
Theory, Applications and Systems, Washington DC (2008)
25. Zhu, X., Ramanan, D.: Face detection; pose estimation, and landmark localization in the
wild. In: Proceedings of Conference on Computer Vision and Pattern Recognition,
Providence, RI, USA, pp. 2879–2886 (2012)
26. Martinez, B., Valstar, M.F., Binefa, X., Pantic, M.: Local evidence aggregation for
regression based facial point detection. IEEE Trans. Pattern Anal. Mach. Intell. 99, 1 (2012)
27. Cao, X., Wei, Y., Wen, F., Sun, J.: Face alignment by explicit shape regression. In: Process
of Conference on Computer Vision and Pattern Recognition, Providence, RI, USA,
pp. 2887–2894 (2012)
28. Tie, Y., Guan, L.: Automatic landmark point detection and tracking for human facial
expressions. J. Image Video Process. 2013, 8 (2013). doi:10.1186/1687-5281-2013-8
29. Mairal, J., Bach, F., Ponce, J., Sapiro, G., Zisserman, A.: Discriminative learned dictionaries
for local image analysis. In: Proceedings of Conference on Computer Vision and Pattern
Recognition, Anchorage, Alaska, pp. 1–8 (2008)
30. Yang, M., Zhang, L., Feng, X., Zhang, D.: Fisher discrimination dictionary for sparse
representation. In: Process of International Conference on Computer Vision, Barcelona,
Spain, pp. 543–550 (2011)
31. Salakhutdinov, R., Torralba, A., Tenenbaum, J.: Learning to share visual appearance for
multiclass object detection. In: Process of Conference on Computer Vision and Pattern
Recognition, Colorado Springs, USA, pp. 1481–1488 (2011)
32. Torralba, A., Murphy, K.P., Freeman, W.T.: Sharing visual features for multiclass and
multiview object detection. IEEE Trans. Pattern Anal. Mach. Intell. 29(5), 854–869 (2007)
33. Black, M., Yacoob, Y., Jepson, A., Fleet, D.J.: Learning parameterized models of image
motion. In: Process of Conference on Computer Vision and Pattern Recognition, San Juan,
Puerto Rico, pp. 561–567 (1997)
34. Wang, H., Ullah, M.M., Kläser, A., Laptev, I., Schmid, C.: Evaluation of local
spatio-temporal features for action recognition. In: Proceedings of British Machine Vision
Conference, London, UK, p. 127 (2009)
35. McDuff, D., el Kaliouby, R., Picard, R.W.: Crowdsourcing facial responses to online videos.
IEEE Trans. Affect. Comput. 99, 456–468 (2012)

View publication stats