Face Annotation using Co-Relation based Matching for Improving Image Mining in Videos

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 04 Issue: 06 | June -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 2579
Face Annotation Using Co-Relation Based Matching For Improving
Image Mining in Videos
Vijay W. Pakhale1, Mayank Bhargav2
1Student, Patel College of Science and Technology, RGTU, Bhopal
2Assistant Professor, Patel College of Science and Technology, RGTU, Bhopal
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In today’s world, Sharing of information online
between friends have boosted the requirement of better
Image retrieval techniques. In online social networking,
sharing of photos and videos in the form of information has
increased the need of Face Annotation. The face annotation
means something particular about the Facial features. Face
Annotation is a bit complex in the videos. Since face
annotation provides the special or the to the point
information about the targeted part of face or the facial
feature, it got the special attention of the programmers.
In this paper, the problems of the face recognition
techniques have been detected andsolvedthem. Thisisdone
using the features correlation based matching techniques.
We observed that the proposed system is quite effective
compared to the others.
Key Words: Image Retrieval, Content-Based Image
Retrieval (CBIR), Unsupervised Label Refinement (ULR),
Search Based Face Annotation (SBFA), Viola Jones
Algorithm.
1.INTRODUCTION
From the popularity of cameras and social media tools,
images and videos gain much more interest of the users. We
can capture videos and images at any time with digital
camera and post it onto the internet using some social
networking systems. Also these images contain some meta-
information like GPS co-ordinates, date-time or quality of
image, which are may be automatically put by our digital
camera [1]. From these images, a largenumberofimagesare
human facial images on internet/social networks. Many
people like to tag their facial images by their names or some
text. But many people does not tag images properly which
leads to inaccurate information about that face. These
images are weakly labeled and their meta-information is
known as weakly labeled data [4]. Numbers of techniques
are used in networking for refining the weakly labeled data,
for getting correct information.
1.1 Biometric Systems:
A person can be generally identified by his face in social
environment. In the same way in machine environment a
person is identified by the Biometric features. Every person
has some unique biometric features and these features are
detected by the biometric system [2][11]. Thesefeatures are
compared with other person’s biometric features to decide
the uniqueness of the first person. In same way, the face
recognition systems perform their function to detect the
faces among other faces. The advantages of these features
are uniqueness and can’t be forgotten. The examples of the
biometric functions are fingerprints, eye retinal featuresetc.
facial features can also be stated as the biometrics.
1.2 Unified Label Refinement:
For weakly labeled facial images collected from internet, an
ULR is an effective method to fix the errors in annotations. It
can effectively correct images those are incomplete and full
of noise. By using ULR in algorithms we can improve
scalability, efficiency and performance of face detection
systems [4][13].
But ULR also has some limitations according to face
annotation. First, it consumes time and collecting a huge
amount of human-named training facial imagesisexpensive.
Next, it is normally hard to generalize the models when new
training data or new persons are added, in which an
intensive retraining process is usually required [2][6].
Frequently the annotation/recognition performance will be
poor if the number of persons/classes is very huge.
1.3 Content-Based Image Retrieval:
Textual information about images can be easily searched
using existing technology, but this requires humans to
manually describe each image in the database. Initial CBIR
systems were developed tosearchdatabasesbasedonimage
color, texture, and shape properties. An image isretrievedin
CBIR system by adopting several techniquessimultaneously
such as Integrating Pixel Cluster Indexing, histogram
intersection and discrete wavelet transform methods [3]. It
can compare content using image distance measures and by
some query techniques like semantic retrieval, relevance
feedback, iterative machine learning, etc.
There are some limitations according to CBIR techniques.
The learning of image similarity, the interaction with users,
the essentiality for databases, the problem of evaluation,the
semantic gap with image features, and the understanding of
images are degrade this technique [7][8].

2. LITERATURE SURVEY
In this section we discuss about previous surveys which are
related to this paper. We foundmanypreviouspapersandby
examining those we describe our method and how it is
better than those others. Some of these are given below:
A.W.M. Smeulders, M. Worring, S. Santini, A. Gupta, and R.
Jain, “Content-Based Image Retrieval at the End of the Early
Years,” IEEE Trans. Pattern Analysis and Machine
Intelligence, vol. 22, no. 12,pp. 1349-1380, Dec. 2000:
It uses image retrieval systems for image processing and
retrieves texture, colour and local geometry of the images
[3]. It considers some features of image retrieval, likesalient
points, shape and object features, their structural
combinations and global and accumulative features. For
each of that features it can review the similarity of pictures
and objects presents in that images. So that, it provides
interaction by considering close connection of types and
feedbacks means for those images. Also author presents
field’s driving force, computer vision heritage and influence
of computer vision for concluding their system.
T.L. Berg, A.C. Berg, J. Edwards, M. Maire, R. White, Y.W. Teh,
E.G. Learned-Miller, and D.A. Forsyth, “Names and Faces in
the News,” Proc. IEEE CS Conf. Computer Vision and Pattern
Recognition (CVPR), pp. 848-854, 2004:
By applying appropriate discriminate coordinates, we can
cluster these face images with associated set of names,
breaks the ambiguities between labeling and identify faces
which are incorrectly labeled, from set of images [5]. After
that it use merging procedure to identify variants of name
applied to same individual from news pictures. It can also
used as process that removes and correct noisy
unsupervised data from any network.
X.-J. Wang, L. Zhang, F. Jing, and W.-Y. Ma, “Anno Search:
Image Auto-Annotation by Search,” Proc. IEEE CS Conf.
Computer Vision and Pattern Recognition (CVPR),pp.1483-
1490, 2006:
In their proposed system at least one keyword is required
which is accurate for enabling text-based search, then to
retrieve visually similar images by content-based search.
After that, from titles and surrounding texts and URLs of
these images, annotations are mined. It ensures high
efficiency with visual features which are mapped by some
hash codes [8]. The search process is speed-up bythesehash
codes. They choosereal web imagesforexperimental results.
Z. Cao, Q. Yin, X. Tang, and J. Sun, “Face Recognition with
Learning-Based Descriptor” IEEE Conf. Computer Vision and
Pattern Recognition (CVPR), pp. 2707-2714, 2010: They
present Unsupervised Learning techniques as encoder for
invariance of training images. For getting compact face
descriptor they apply PCA technique and simple
normalization. The result of that technique produces high
discriminative and easy to extract learning-baseddescriptor
[13]. They also propose pose-adaptivematchingmethodsfor
handling large pose variations of matchingfacepairbyusing
pose-specific classifiers for some pose combinations. They
also achieved 84.45% recognition rate on the different
datasets like Labelled Face in Wild Benchmark.
2.1 Limitations of Existing Systems:
We have many existing systems which resemble our work,
but all these existing systems have more or less limitations
in their fields. These limitations cause poor result or
inconveniency as we get output from any existing
application which uses these existing concepts. Some of
these limitations from existing systems are consideredhere:
2.2 Content Based Image Retrieval:
i. In this technique, the tagging of the image is done
efficiently. But the metadata is to be entered
manually for the creation of database and
tagging [11].
ii. The Systems using this technique is not efficiently
good enough to have a better interaction with
users.
iii. The complexity of statistical evaluation gets
increased [10].
iv. The manual tagging of the image features creates
the semantic errors.
v. The accuracy of the system is compromised

2.3 Search-Based Face Annotation:
i. In this technique, mostly the annotation is
performed on the images which are Famous
like celebrities. So it’s generalization of the
models is difficult when new data or new
persons are considered [9].
ii. Due to the long process of matching, the technique
consumes the time
iii. It is an expensive task to collect the huge amount of
face features data primarily required here.
iv. Classifier defined classes have the inverse effect on
the performance of the systems. The numberof
classes increases, the performance decreases
[12].
3. PROPOSED SOLUTION
Here we define our algorithm for face annotation which is
accurate and less time consuming with respect to other
systems. In our proposed solution first we get our system
raw data from online sources in the form of videos then we
divide these videos in several images. After that by running
our algorithm we retrieve those images which are best
images for providing the resource data to the system
accurately. After selecting images we perform feature
extraction operations for collecting the features of the faces
which are present in those images. These features can be
compared with other facial features presentinourdatabase.
These matched facial features can further define the exact
person present in a particular image. After that our system
can automatically tagged that images with those data
through which another system can also gets the accurate
information.
Image Retrieval from Video: In our proposed system, the
videos which are uploaded by the users are observed. The
images are retrieved from them using frames separation.
The images which are not proper are rejected. The proper
ones are considered for further process.
Measuring Facial Features: The next step is measurement of
the facial features of the proper images chosen. In this
proposed solution we measure several different facial
features of the image. Like,
i. Width of an eye
ii. Height of an eye
iii. Width of nose
iv. Height of nose
v. Width of mouth
vi. Height of mouth
vii. Distance between two eyes
viii. Distance between left eye to nose
ix. Distance between right eye to nose
x. Distance between left eye to mouth
xi. Distance between right eye to mouth
xii. Distance between nose to mouth
To measure these features, we have to consideronlytheface
region of every frame of the image. For this very purpose,
grey scaling of the image is performed which helps in the
detection of these features.

1. Database Creation: Every image which has undergone
the measurement of features is stored in the database.
This database is specially formed to assign the feature
statistics for the particular image. The Database is also
accessed while comparison process. The statistical
information stored in database is reused in face
recognition process.
2. Comparing Features within Images: By observing the
images we get the individual facial features.Thefeatures
of the images stored in the database are compared with
these observed facial features. The relative statisticsare
considered for matching or the non-matching of the
images. For example, the distance between the eyes of
one image is compared with the distance between the
eyes of the other image.
3. Face Recognition from Database: In the face recognition
step, the comparison is checked within the system for
the maximum matching of the respective features.
Through this, the system knows the differences of the
face present in the images. Whentheobservedmatching
ratio is high, the possibility of those two faces to be
similar gets higher. In our proposed method for face
recognition the accuracy of the matching ratio is higher
than the other algorithms.
4. Face Annotation: Thisabovementioned wholeprocessis
the backbone for the face annotation. The statistical
information we get through this entire process is the
note or the explanation added to the image. This fulfils
the necessary conditions of face annotation.
4. RESULT ANALYSIS
To analyze the results ofthe proposedtechnique,theprocess
is to be carried out is two main parts. These two parts are
divided on the basis of functionality performedinrespective
parts of the system. The second part uses the information
which is generated by first part.
To create the database, the image or video is input in the
system. If the input is video, the proper usable images are
retrieved from the video by dividing it in the frames of
images. Face detection process is carried out on the image
using Viola-Jones algorithm. The grey scale of the detected
face also called the encoded face is obtained. This grey scale
image helps in providing the proper locations of the facial
features like eyes, nose etc. Once the facial features are
extracted from the image, the computational analysis of the
features is begun. The twelve parameters mentioned above
are measured. Identification is provided to the framesorthe
image. The measurements tagged with image are stored in
the database.
Next part of the process is to evaluate the system. The image
or video to be verified against the database created earlier.
The video is again divided in the frames and exposed to
Viola-Jones algorithm for face detection. Whereastheimage
is directly fed into system. The process of grey scaling and
facial features extraction is carried out as above. The
computation of the features is done. Now the related
parameters are compared with theparametersoftheimages
saved in the database. The images from database with
highest matching relative parameters are shown as output.
In our System, we created the database using both videos
and images of 4 different persons. The total number of the
images fetched by system in the database was 22. Then for
the evaluation of the system we chose one video of a person
with 20 frames. After the evaluation performed by the

system, it gave 6 most matched images from database with
the percentage accuracies as 99.77%, 99.85%, 99.85%,
99.98%, 99.98% and 99.97% respectively. This resulted in
the average percentage of accuracy as 99.90% for our
proposed system. The delay factor in this process for
comparison given by our system is 0.6848 seconds for 22
frames. This gives the average delay of 0.03113 seconds,
which is less than the existing systems.
5. CONCLUSION
Through our proposed technique, we have proved that the
face annotation technique by measurementoffacestructural
features is quite adequate. The accuracy of the system has
been increased. The performance factor has been increased
with respect to time. Due toimprovedaccuracyanddecrease
in the delay, the reliability of our system increases. The
systematic database used in our system helped to achieve
the desired performance. In this technique of face
annotation, the drawbacks of some existing face annotation
systems have been eliminated to some extent.
This research indicates a new direction for the findings of
asymmetric systems to eliminate the dependence on the
passwords and smart cards etc. This face annotation
technique may be a useful tool for the improvement of
identity recognition parameter in online social networking.
6. REFERENCES
1. S. Satoh, Y. Nakamura, and T. Kanade, “Name-It:
Naming and Detecting Faces in News Videos”, IEEE
MultiMedia, vol. 6, no. 1, pp. 22-35, Jan.-Mar. 1999.
2. Jianke Zhu, Steven C.H.HoiandMicheal R.Lyu,“Face
Annotation Using Transductive Kernel Fisher
Discriminant”, IEEE Trans. Multimedia, vol. 10, no. 1,pp.
86-96, Jan. 2008
3. A.W.M. Smeulders, M. Worring, S. Santini, A. Gupta,
and R. Jain, “Content-Based Image Retrieval at the End of
the Early Years,” IEEE Trans. Pattern Analysis and
Machine Intelligence, vol. 22, no. 12,pp.1349-1380,Dec.
2000
4. D.-D. Le and S. Satoh, “UnsupervisedFaceAnnotation
by Mining the Web,” Proc. IEEE Eighth Int’l Conf. Data
Mining (ICDM), pp. 383-392, 2008
5. T.L. Berg, A.C. Berg, J. Edwards, M. Maire, R. White,
Y.W. Teh, E.G. Learned-Miller, and D.A. Forsyth, “Names
and Faces in the News,” Proc. IEEE CS Conf. Computer
Vision and Pattern Recognition (CVPR), pp. 848-854,
2004
6. Ms. Malashree Patil, Prof. Jyoti Neginal, “A Robust
Face Annotation Method by Mining Facial Images”,
IJARCET, May 2015.
7. Dayong Wang, Steven C.H. Hoi, Ying He and Jianke
Zhu, “Mining Weakly Labeled Web Facial Images for
Search-Based Face Annotation”, IEEE, Jan 2014.
8. X.-J. Wang, L. Zhang, F. Jing, and W.-Y. Ma, “Anno
Search: Image Auto-Annotation by Search,”Proc.IEEECS
Conf. Computer Vision and Pattern Recognition (CVPR),
pp. 1483-1490, 2006
9. Steven C.H. Hoi, Dayong Wang, Yeu Cheng, Elmer
Weijie Lin, Jianke Zhu, Ying He, Chunyan Miao, “FANS:
Face Annotation by Searching Large-scale Web Facial
Images”, IW3C2, May 2013.
10. M. Kan, S. Shan, H. Zhang, S. Lao and X. Chen, “Multi-
View Discriminate Analysis”, in proc, 12th Eur. Conf.
Comput. Vis., vol. 7572. Oct. 2012, pp. 808-821.
11. G. Guo and X. Wang, “Kinship Measurement on
Salient Facial Features”, IEEE Trans. Instrum. Meas., vol
61, no. 8, pp. 2322-2325, Aug. 2012.
12. Y. Yang, Y. Yang, Z. Huang, H.T. Shen,andF.Nie,“Tag
Localization with Spatial Correlations and Joint Group
Sparsity,” Proc. IEEE Conf. Computer Vision and Pattern
Recognition (CVPR), pp. 881-888, 2011.
13. Z. Cao, Q. Yin, X. Tang, and J. Sun, “Face Recognition
with Learning-Based Descriptor” IEEE Conf. Computer
Vision and Pattern Recognition (CVPR), pp. 2707-2714,
2010

Face Annotation using Co-Relation based Matching for Improving Image Mining in Videos

More Related Content

What's hot (17)

Similar to Face Annotation using Co-Relation based Matching for Improving Image Mining in Videos (20)

More from IRJET Journal (20)

Recently uploaded (20)

Face Annotation using Co-Relation based Matching for Improving Image Mining in Videos