Date: 22.3.2015
INTERNATIONAL CONFERENCE ON EMERGING TRENDS IN ENGINEERING RESEARCH
AbstractDay by day there is a very steep increase in digital media devices. So the number of photos shared on web is also
high. Among those shared photos only some of the images are uploaded with names properly. Many of them are untagged. To
overcome this, faces need to be annotated automatically. It is a challenging task to annotate an image by searching weakly
named images from internet based on detecting and recognizing the faces. Some approaches are used to name images. Those
approaches include Search-based face annotation (SBFA), which searches the available untagged images on the web (WWW).
SBFA approach aims to effectively assign names to noisy and incomplete facial images. Another approach to assign name is
unsupervised label refinement (ULR), which filters the names of facial images on the web. And further to improve the scalability, a clustering-based approximation algorithm is used.
Keywords Facial annotation, content-based image retrieval, machine learning, label refinement, web facial images, weakly labeled, search based face annotation, Non-tagged image.
I.INTRODUCTION
With the tremendous increase in the development of digital devices such as digicam and scanners [1], a photograph
taken by humans also increases. They share these photographs through social websites. But they are not shared with
names. It is a tedious task to organize the photographs by annotating those using commercial products [1].A new technique to retrieve images is done using content-based image retrieval (CBIR) [2]. Content-based image retrieval is the
method of retrieving images by using their content (QBIC) [2]. Content-based visual information retrieval (CBVIR) [2]
is the method to search images from large databases, thus solving image retrieval problem. The model-based face annotation technique has some limitations [3]. The major limitation is, it took more time to collect images from database and
also it requires more cost to do this. Then, it is more difficult for generalizing the technique when new data or new persons are added. Finally, achieves poor scalability when number of persons increases [4]. To overcome the above limitations, images uploaded on the web must be automatically annotated. Automatic facial annotation is important technique
which automatically gives name of relevant person. This technique has more benefits in many real world applications.
Many users upload their photos on social websites such as face book, twitter, etc. These photos are not annotated properly. Likewise, there are large numbers of untagged images on social websites. To annotate those images, Search Based
Facial Annotation paradigm is used [3]. Instead of using model-based face annotation approaches, we use search-based
face annotation (SBFA) paradigm [4].This paradigm uses content-based image retrieval (CBIR) technique [4] to automatically name the images. The SBFA framework is based on data and does not depend on models. The main objective of
SBFA is to assign names for a given query image correctly [4].From a database that contains only the untagged images,
we shortlist top K most images that are similar and then by getting votes on the names associated with those top K images, we annotate them[3], [4]. But the limitation with SBFA paradigm is to effectively use the advantage of shortlisted
images and their weak names. The effectiveness can be achieved by developing a search-based face annotation scheme.
Without any manual efforts, we pull out the names homogeneously from the strength less data using unsupervised label
refinement (URL) scheme. In order to improve the efficiency and to perform this for a large number of images, clustering-based approximation (CBA) algorithm is used [3], [4].
II.RELATED WORKS
Different researches have been performed on face annotation in searching non tagged facial images which are present
over internet. Human name is given as input query and we need to get consistent facial images as output. One group focuses on recognizing and verifying the faces, which exists in classical research problems in recognition of pattern. Next
group focuses on generic image annotation. In classical method, they use some object recognition techniques to identify
the similarities and relationship between image and the input text. For example, with a label similarity graph and random
walk principle [5], Wang et al. proposed to purify the results obtained by traditional technique. Similarly, with an iterative label propagation scheme [6], Pham et al.[6] recommended to name the untagged images that exists in video frames.
Another group focuses on annotating the faces of personal/ family/social photos. Personal photos contain personal
names, things that happen in society, geographical information and time when the photo is captured which are associated
with the images. For example, commercial products such as Apple iPhoto, Google Picasa, Microsoft EasyAlbum[7],
[8]and Facebook face auto tagging solution achieve a good naming results.
IAETSD 2015
http://www.iaetsd.in/proc-22.3.2015.html
ISBN: 978-15-086565-24
Date: 22.3.2015
INTERNATIONAL CONFERENCE ON EMERGING TRENDS IN ENGINEERING RESEARCH
Another group focuses in searching non-tagged facial images on the web to annotate them. Ozkan and Duygulu [9]
recommended a graph-based model to find the sub-graph with more density as a result for the input image. Berg et al.
[10] recommended a possibility model aggregated with clustering algorithm to measure the correlation between the images and the names associated with images. For the images and its corresponding caption in the same document, Guillaumin et al. [11] recommended to repeatedly rejuvenate the assignment based on a least cost matching algorithm.
Last group focuses on clarifying facial images, which aims to analyze inaccurate web images for identifying faces[10].
For a large and preliminary set, a consistency learning method [12] is applied for searching text-image in sequence or
simultaneously on the web.
III.CONTENT BASED IMAGE RETRIEVAL
"Content-based" means searching the figure based on their content instead of searching the image based on metadata
or descriptions associated with the image [2]. "Content" may be pixel, structure, contrast, or any other information that
appears on the image itself [2].
Figure 1. Reveals the flow of the proposed scheme of content based image retrieval
It consists of the following steps:
1. Collecting data;
2. Detecting and extracting the features of images;
3. Indexing the extracted features of images using high dimensions;
4. Filtering the images from database;
5. Retrieval of similar faces; and
6. Annotating the retrieved similar faces by majority voting using filtered names.
Before the analysis phase of a naming task, the first four steps are performed and at the time of analysis phase of a
naming task, the last two steps are performed [4]. The first process is the accumulation of data about facial images. Using
web search engine, images of the person are collected according to a name list from Google [4]. As a result, this step
contains facial images with some human names. These facial images do not contain the right human name. So they are
called as inaccurately named images. The second process is preprocessing facial images on web to obtain information
related to the face such as finding and arranging, region separation, and component illustration. Unsupervised face
alignment technique [13] is used for finding and adjustment of faces. The third step is tabulating the obtained components of the faces. This is done by locality sensitive hashing (LSH)[14], a very familiar and powerful high-dimensional
indexing technique to fetch the similar faces only. In the last three phases [4], first we fetch group of most similar faces
i.e. top k images from the previously tabulated facial database. By majority voting [4] approach, the facial image is defined with a name by combining the set of names associated with these top K images.
IAETSD 2015
http://www.iaetsd.in/proc-22.3.2015.html
ISBN: 978-15-086565-24
Date: 22.3.2015
INTERNATIONAL CONFERENCE ON EMERGING TRENDS IN ENGINEERING RESEARCH
IAETSD 2015
http://www.iaetsd.in/proc-22.3.2015.html
ISBN: 978-15-086565-24
Date: 22.3.2015
INTERNATIONAL CONFERENCE ON EMERGING TRENDS IN ENGINEERING RESEARCH
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
IAETSD 2015
L. Chen, Automated annotation of human faces in family albums, pp. 25, 2003.
J. Eakins, Content based image retrieval, 2014.
S. Pa. Rahul Jadhav, A survey on mining weakly labeled web images, IEEE Trans. Knowl. Data Eng., vol. 26, no. 1, pp.
166179, 2014.
D. Wang, S. C. H. Hoi, Y. He, and J. Zhu, Mining weakly labeled web facial images for search-based face annotation, IEEE
Trans. Knowl. Data Eng., vol. 26, no. 1, pp. 166179, 2014.
T. Pagerank, C. Ranking, and B. Order, 1. Introduction and Motivation 2. A Ranking for Every Page on the Web, pp. 117,
1998.
B. Heverlee, Naming persons in news video with label propagation Phi The Pham , Marie-Francine Moens Katholieke Universities Leuven Department of Computer Science Tinne Tuytelaars Katholieke Universities Leuven ESAT - PSI Kasteelpark
Arenberg 10, 2010.
J. Cui, F. Wen, R. Xiao, Y. Tian, and X. Tang, EasyAlbum: An interactive photo annotation system based on face clustering
and re-ranking , pp. 367376, 2007.
Y. Tian, W. Liu, R. Xiao, F. Wen, and X. Tang, A face annotation framework with partial clustering and interactive labeling,
2007 IEEE Conf. Comput. Vis. Pattern Recognit., pp. 18, Jun. 2007.
D. Ozkan and P. Duygulu, A graph based approach for naming faces in news photos, Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit., vol. 2, pp. 14771482, 2006.
T. L. Berg, A. C. Berg, J. Edwards, M. Maire, R. White, Y. Teh, E. Learned-miller, and D. A. Forsyth, Names and faces in the
news.
M. Guillaumin, T. Mensink, J. Verbeek, C. Schmid, and C.S.Automatic, Automatic face naming with caption-based supervision, 2011.
J. Yagnik, H. Adam, and D. Bau, Large scale learning and recognition of faces in web videos, 2008 8th IEEE Int. Conf. Autom. Face Gesture Recognit., no. 1, pp. 17, Sep. 2008.
J. Zhu, K. U. Leuven, and S. C. H. Hoi, Unsupervised face alignment by robust nonrigid mapping, no. 1.
W. Dong, Z. Wang, W. Josephson, M. Charikar, and K. Li, Modeling LSH for performance tuning, Proceeding 17th ACM
Conf. Inf. Knowl. Min. - CIKM 08, p. 669, 2008.
Z. Wu, Q. Ke, J. Sun, and H. Y. Shum, Scalable face image retrieval with identity-based quantization and multireference reranking, IEEE Trans. Pattern Anal. Mach. Intell., vol. 33, no. 10, pp. 19912001, 2011.
http://www.iaetsd.in/proc-22.3.2015.html