IRJET- Segmenting, Multimedia Summarizing and Query based Retrieval of News Story from News Broadcast

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 04 | Apr 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 728
Segmenting, Multimedia Summarizing and Query Based Retrieval of
News Story from News Broadcast
Mr. N. V. Bhalerao1, Dr. Mrs. S. S. Apte2, Dr. Mrs. A. R. Kulkarni3
1 ME Student, Dept. of Computer Science and Engineering, WIT, Maharashtra, India
2 Professors, Dept. of Computer Science and Engineering, WIT, Maharashtra, India
3 Professors, Dept. of Computer Science and Engineering, WIT, Maharashtra, India
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract- This project builds a system to summarize and
retrieve the broadcast news at multimedia level. This project
combines anchor person based story boundary detection and
text summarization system to build multimedia news
summary and news extraction system. Broadcast news are
captured both in video/audiowithaccompanyingtranscriptin
text format. Summarized individual news story video clip
according to textual summary can be retrieved by user query.
Lexical chain text summarization technique is used to
summarize individual news story transcript. The summary is
in multimedia format including video, audio, and text.
Key Words: Multimedia Summarization, Retrieval
System, Story Boundary Detection,TextSummarization,
Lexical Chain, News Broadcast.
1. INTRODUCTION
Broadcast news video has been playing all the time
more vital role in our everyday life. As a kind of
primary multimedia resources, broadcast newsvideos
are regularly accessed by a large number of people all
over the world. The broadcast news video has several
distinct characteristics that are quite different with
other types of videos. For instance, a story unit always
accompanies with one to several descriptive caption
texts. Anchor persons usually appear at the beginning
of a story unit. Thus, the broadcast news video can be
regarded as a kind of semi-structured multimediadata
that contains informative clues for parsing itself into
semantic story units.
Broadcast news videos are regularly accessed by a
large number of people all over the world. However,
despite its popularity, finding video of a particular
person’s interest is still by no means easy for several
reasons. First, broadcast news videoarchivescontaina
large number of historical videos. Facing the
overwhelming amounts of videos, people are always
difficult to locate the videos they are interested in, and
thus ‘‘lost in TV program space’’. Second, the latest
broadcast newsvideosaredelivereddailyatfixedtime,
people sometimes may not conveniencetowatchthem
due to time conflict. In addition, there are cases that
two videos a person both interested in are deliveredas
the same time. Third, instead of watching the whole
video, sometimes what people want to watch is only a
particular segment of interest.
To efficiently manipulate and manage the increasing
broadcast news videos, a key technique is by
segmenting the whole video into meaningful and
relatively independent video clips each depicting only
one story, i.e., story units.
The aim is to build broadcast news summarization
system. It takes YouTube news broadcast, analysis the
content to identify news stories. Content of each story
is summarized and important keywords are extracted.
This data is to be stored in a database, and news
Retrieval system is to be implemented which let users
to explore for any part of news in the database. Such a
system have improvement over other search engines,
as we use summarization methods to focus the most
important information to the user, facilitatinghim/her
to find the story he/she is eyeing for in a smaller time,
when compared to an ordinary text based search
engine.
1.1 Aim of Project
 Story segmentation – in order to identify story
boundaries from 30 minutes worth of video
and transcript(subtitles).(E.g.tofindthatthere
are 10 individual pieces of news in a
RajyaSabha TV news bulletin at night). A
technique suitable to solve this problem is
story segmentation using anchor person. It
identifies the current story and detects
when/where the story changes. It helps to
detect thestoryboundariesandidentifiesstory
frames. There are known methods to carry out
video segmentation but all of them suffer from

certain limitations, and this problem has been
proved to be difficult to solve. We try to
implement story segmentation with the use of
anchor person frame to segment the videointo
news stories. Anchor person usually appear at
the beginning of every news story in news
bulletin.
 Story summarization–providestheuserwitha
short summary of the story both in text and
video. This lets the user to choose whether a
video excerpt returned by the search engine is
related to the subject he/she is searching for.
We apply a text summarization technique
called ‘Lexical Chain’ to summarize the text.
Video summary of story is generated using
summarized text.
 Keyword based news retrieval system-apart
from a summary of the original story, we could
also try to identify importantkeywordssuchas
“Delhi”, “Prime Minister” or “blast” from news
transcript for each story segment. Such
keywords, if recognized correctly, would
contribute incredible amount to specify what
the topic of the story is about. Based on
frequency of occurrences of matching
keywords user can select stories of interest.
Video analysis techniques are used to story
segmentation,extractingprominentfeaturefromvideo
and summarizing the story video as per text summary.
We have used natural language processing technique
and tool (WordNet –lexical database) to extract
important keyword and summarize the text.
2. LITERATURE RIVIEW
Mark T. Maybury and Andrew E. Merlino has
implemented and extracted “Summaries for Broadcast
News”. They have used algorithm for summarization,
key phrase extraction and story segmentation, and key
frame extraction generate the summarized video.
Marcus J. Pickering, StefanM. Rüger has focused “video
search engine using dual-media segmentation”. They
implemented an algorithm which uses the audio track
for identifying meaningful scene breaks. This work is
related to web-based video search engine that is
implemented using broadcast news, and the main part
of implementation is story boundary detection.
Kuan-Yu Chen, Shih-Hung Liu has implemented
“ExtractiveBroadcastNewsSummarizationLeveraging
Recurrent Neural Network Language Modeling
Techniques” their workinthispapermainlyfocusedon
use of recurrent neural network language modeling
(RNNLM) framework for extraction as well as
summarization of broadcast news.
MarkT.Mayburymainlyfocusedon“DiscourseCuesfor
Broadcast News Segmentation” they describe analysis
ofabroadcastNewscorpus,andfocusedoninformation
extraction techniques, and finally its computational
implementation and evaluation in the Broadcast News
Navigator (BNN) for achieving browsing, retrieval, and
summarization of news video.
Warren Greiff, Alex Morgan, Randall Fish, Marc
Richards, Amlan Kundu presented “Fine-Grained
Hidden Markov Modeling for Broadcast-News Story
Segmentation”. The News broadcasts are divided into
story segments by using Hidden Markov Model. Model
topologyandthetextualfeaturesusedtogetherwiththe
non-parametricapproximationtechniquesforobtaining
estimates for both transition and observation
probabilities. Visualization approaches developed for
the examination of system performance.
Kathleen McKeown and Dragomir R. Radev have
established a model for “Generating Summaries of
Multiple News Articles”. They have offered Natural
Language system for summarization of a sequence of
News articles on the same event.
Hemant Misra, Frank Hopfgartner, Anuj Goyal, P.
Punitha, and Joemon M. Jose are focused on “TV News
Story Segmentation based on Semantic Coherence and
Content Similarity”. They have assessed two
methodologies, one using video stream and the other
using close-caption text stream, for segmenting TV
newsintostories.Thesegmentationofthevideostream
into stories is achieved by detecting anchor person
shots and the text stream is segmented into stories
using a Latent Dirichlet Allocation (LDA) based
approach.
Regina Barzilay,Michael Elhadad have explored one
method to summarize original text by using the model
ofthetopicprogressioninthetextresultingfromlexical
chain. They offered new algorithm to compute lexical
chain in the text.
Zechao Li, Jinhui Tang ,Xueming Wang, Jing Liu ,
Hanqing Lu havedevelopednewmethodofmultimedia
news summarization for searching results on the

Internet, which finds the essential topics amongquery-
related news information and threads the news events
within each topic to generate a query-related brief
summary. They used HLDA to topic structure from
query associated news document. And time influenced
maximum spanningtreealgorithmissuggestedtoform
condensed summary of parent topic.
Peter Bell, Catherine Lai, Clare Llewellyn, Alexandra
Birch, Mark Sinclair defined an end-to-end system for
processing and browsing audio news data. Their fully
automated system carries together recent research on
audio scene analysis, speech recognition, and
summarisation, named entity detection, geo location,
and machine translation.
Jia-Yu Pan, Hyungjeong Yang, and Christos Faloutsos
recommended multi-modal story-oriented video
summarization (MMSS). MMSS discovers association
between information of different modalities which
givesexpressivestory-orientednewsvideosummaries.
MMSS can also be applied for video retrieval.
Ichiro Ide, Ye Zhang, Ryunosuke Tanishige, Keisuke
Doman,Yasutomo Kawanishi, Daisuke Deguchi, and
HiroshiMurase putforwardamethodforsummarizing
asequenceofnewsvideosconsideringthesteadinessof
both auditory and visual contents. The suggested
technique first selects key-sentences from the auditory
contents (Closed Caption) of each news story in the
sequence, and then picks a shot within the news story
whose “Visual Concepts” identified from the visual
contents are the most consistent with the key-phrase.
Finally,theaudiosegmentmatchingtoeachkey-phrase
is coincided onto the selected shot, and then
concatenated to generate a summarized video.
Mi-mi LU, Lei XIE, Zhong-hua FU, Dong-mei JIANG and
Yan-ningZHANGstudiedhowtoassimilatemulti-modal
features for story boundary detection in broadcast
news. The uncovering problem is expressed as a
classification task, i.e., classifying each candidate into
boundary/non-boundary based on a set of features.
They used a varied collection of features from text,
audio and video modalities: lexical features capturing
the semantic shifts of news topics and audio/video
featuresreflectingtheeditorialrulesofbroadcastnews.
BailanFeng,ZhinengChen,RongZheng,BoXuproposed
a new uniﬁed video structure parsing method, named
multiple style exploration-based news story
segmentation (MSE-NSS), to segment broadcast news
videos into semantic story units. In MSE-NSS, they ﬁrst
explore the suitable methods to explore various kinds
of style information intrinsic in broadcast news videos,
comprising temporal style inferred from caption texts,
boundary style signified by a affluence of multimodal
visual–audiofeatures,andstructuralstyleknownasthe
spanning duration of story units. The task of story unit
segmentation is accomplished through the following
three steps: temporal style-based pre-location,
boundary style-based description, and boundary-
structural style-based segmentation. Parallel to this, a
news-orientedbroadcastmanagementsystem—NOBMs
is implemented on top of the proposed MSE-NSS.
Jae-Gon Kim, Hyun Sung Chang, Kyeongok Kang,
Munchurl Kim, Jinwoong Kim, Hyung-Myung Kim
suggested a newmethodforsummarizinganewsvideo
based on multimodal analysis of the content. The
suggestedmethod exploits the closedcaption(CC)data
to locate semantically meaningful highlights in a news
video and speech signalsin anaudio streamtoalignthe
CC data with the video in a time-line. Then, the
extracted highlights are described in a multilevel
structureusingtheMPEG-7SummarizationDescription
Scheme (DS).
3. METHODOLOGY
3.1 Multimedia Summarization and Retrieval of
News Broadcast
We define the system by identifying the 5 main
components of the system; each of them will be
discussed in full details in the following sections:
Fig-1: Main Parts of System
News
Broa
dcast
,
Trans
cript
Story
Identif
ication
Text
Summa
rization
Video
Summar
ization
Keyword
Based
Retrieval
System

News Broadcast and Transcript: - Input to this
system includes news Broadcast video and Transcript.
News Broadcast is broadcasted on RajyaSabha TV at 9
pm as news bulletin. We downloaded news broadcast
and transcript from YouTube.
Story Identification: - News Broadcast video is
analyzed processed and stories boundaries are
detected, news stories are identified. Anchor Person
based story boundary detection is used to detect story
boundaries and identify stories. The news video and
transcript issegmentedaccordingtothedetectednews
story boundary.
Text Summarization: - The most vital information
condensed and extracted to produce news abstracts,
i.e. a summarized passage for each news story. Natural
language processing technics are used to summarize
newsstorytranscript.Lexicalchaintextsummarization
technique is used to summarize each news story.
Video Summarization: - Each news story video is
summarized as per text summary generated by using
lexical chain text summarization algorithm. Video
processing technic are used to summarize news story.
Storyframescorrespondingtoeachlineinsummarized
story text are extracted to form news summarized
story video.
Keyword based news retrieval system: - TFIDF is
used to extract Important Keywords from each news
stories. This could include a list of locations, persons
mentioned in the news, times/dates of events, etc.
Important keywords from each news story, summary
of each news story are stored in a database, and a
keyword based information retrieval system can be
constructed. Through the use of a search engine news
can be located and extracted efficiently.
3.2 System Architecture
Multimedia summary of news broadcast comprises the
act of taking multimedia stream (news broadcast)
comprising video, audio, and text. The news broadcast
video is divided into distinct news stories segment. By
using anchor person news story boundaries are
identified. And start and end time of each distinct news
story is recorded. By using individual story boundaries
identified using anchor person corresponding news
transcript is segmented into individual news stories.
News transcript consisting of individual newsstoriesis
passed to TF-IDF (term frequency-inverse document
frequency) and natural language processor to identify
important keywords comprising names of persons,
name of locations; events etc. and store it in database.
Thendistinctnewsstoriesaresummarizedusinglexical
chaintextsummarizationtechnique.Summarizednews
story is stored in database. As per the textual summary
generated for individual news story, individual news
video/audio is also summarized. Start time of each line
in news transcript is used to extract story frames in
original news video. AVS file consisting summarized
news story video is formed. Keyword based search is
implemented to get the summary comprising text,
audio/video of news story desired by user. TF-IDF
weighting scheme is used as a central tool in scoring
and ranking a news story’s relevance given by
user query.
Fig. 2 shows the overall architecture of the System.
3.3 Lexical Chain Algorithm
1. Maintains a list of interpretations.
2. Each interpretation consist of a list of lexical
chains
3. Each Chain is a list of pair of nodes...
4. Each pair of node represents a link , and is in
the form:
[$word1, $line1, $word2, $line2]
5. When a link is detected, check existing
chains, and
6. possible append onto chain
7. Otherwise, create new chain with the new
pair.
8. Loop until reach end_line_index
9. At the end of each loop prunes the weak
interpretations
Lexical Chain Text summarization algorithm is
used to summarize individual news story.
There are 3 stages for constructing lexical
chains:
1. Select a set of candidate words
2. For eachcandidateword,findanappropriate
chain relying on a relatedness criterion among
members of the chain
3. If it is found, insert the word in the chain and
update it accordingly

Fig-2: System Architecture
Fig.-2: Overall System Architecture
A lexical chain is created by taking a newtextwordand
finding a related chain for it according to the
“relatedness criteria”. In order to be able to measure
the relatedness criteria of 2 words, i.e. whether 2
words are related to each other even they are different
words, a synonymy dictionary was required.
Candidate Words
All nouns in the story were chosen as candidate words
for lexical chains.
Selecting Strong Chains
After all the nouns in the story have been considered,
the interpretation withthehighestscorewaschosento
represent the story. Then the 3 highest scoring chains
were chosen to be the Strong chains.
Chain score=∑score generatedbyeachlinkinthechain
Scores generated by each link depends on the type of
link.
Extra-strong (between a word and its repetition),
Strong (between two words connected by a WordNet
relation – Synonomy and Hypernomy)
Medium-strong (link between the synsetsofthewords
is larger than one).
In selecting which chain to insert given a candidate
word, extra-strong chains are preferred to strong
relations,whichitselfissimilarlypreferredtomedium-
strong relations.
Summary
Below were the steps used for extracting summary
from Strong Chains :
1. Select the representative word of the chain
2. Extract important sentences to be the summary
Representative word selected
In each chain, the word with the highest occurrence
was chosen to be the representative word.
Extract important sentences.
Once the representative word was chosen, for each
sentence this wordappearsin,wecalculatethescoreof
that function using the weighting function.
Sentence score=∑ (no. of key entity (i) detected) X
weight (type of key entity (i))
4. RESULT AND ANALYSIS
4.1 Story Segmentation
Three newsrecordingswereusedfortrainingdata,and
were manually processed for comparison. Each news
recording was approximately 25 minutes of news
broadcast. The ground truth was identified by
manually viewing the video and the accompanying
transcript, and these werecomparedtotheboundaries
detected by the Anchor person based story
identification algorithm.
News
broadcast
video/audi
o
News
transcript
Detect story
boundaries
using
anchor
person
Summarize
each
segmented
news story
text
Summarize
segmented
news story
video
Identify
segmented
news story
video
Identify
important
keywords
from each
news story
text
Segment
news
transcript
into
individual
stories
Keyword based news story retrieval system
Data
base

Test Results
Test 1
English news bulletin dec 09, 2017 (9 pm)
No. of stories detected-11
No. of real stories-12
False Positive-0
False Negative-1
Percentage of decision correctly made: 1 – 1 / 12 =
91.66%
Test 2
English news bulletin dec 16, 2017 (9 pm)
False Positive-1
False Negative-1
Percentage of decision correctly made: 1 – 2 / 15 =
86.66%
Test 3
English news bulletin jan 06, 2018 (9 pm)
False Positive-0
False Negative-0
Percentage of decision correctlymade:1–0/9=100%
We evaluated the segmentation performanceusingthe
precision Pseg metrics.
Psg = | identified stories | - | wrong stories | / |
identified stories |
Boundaries are correctly detected when a determined
boundary lies within ﬁve seconds of an actual
reference story boundary. Otherwise, the boundary is
considered to be wrong.
Table-1: Precision
News
Broadcast->
English
news
bulletin
dec 09,
2017 (9
pm)
English
news
bulletin
dec 16,
2017 (9
pm)
English
news
bulletin
jan 06,
2018 (9
pm)
Precision(Psg) 0.90 0.84 1.0
The main weakness of the anchor person based story
boundary detection approach seems to be the actual
detection of anchorpersonframe.Wheneverananchor
person frame has been missed, a possible story
boundary will be ignored, hence bring about a drop in
precision. Moreover, stories that do not start with an
anchor person shot will be missed as well, which is a
drawback of our anchor person based story boundary
detection approach.
4.2 Story Summarization
There are no formal methods to evaluate a
summarization algorithm, as 2 human generated
summaries from the same passage could be very
different. Therefore an intrinsic method was used to
evaluate the Lexical Chain algorithm used in this
project.
Story 1 – System Generated results
Beijing unveiled official emblems are for the 2022
Winter Olympics and Paralympic Winter Games the
Winter Olympics emblem was inspired by the Chinese
character dong which means winter the upper half of
the logo was originated from the shape of a speed
skater while the lower part was from the skier the
emblem offer Winter Paralympics was transformed
from Chinese character Fei which means flying Real
Madrid will face a Brazilian club Jaime Oh in the World
Cup of football a final today in last six matches rial has
won four games while two matches ended inadrawon
the other hand out of their six matches Gremiohasa1-
lost - and two games ended in a draw in the semi-finals
Real Madrid beta al Jazeera - one while a cranial beater
patro up 1-0 in the semi-final match to reach the final
of the Club World Cup FIFA's ethics a watchdog
provisionally

banned head of the Brazil's a Soccer AssociationMarco
Polo del Nero for 90 days the FIFA ethics committee
said that del Nero was banned from all international
and domestic soccer activities and could be excluded
for a further 45 days the ban was imposed after a
request from FIFA's the investigatory champ chamber
which is looking into unidentified violations of the
organization's ethics rules that's all in this edition of
fun news but before we go the Eiffel Tower is all set to
attract visitors during the Christmas season with its
transformation as a winter wonderland complete
visitors are being welcomed by a family of a shiny
penguins club chairs and replica of the Eiffel Tower
with mirrors we'll leave you with these are stunning
visuals thanks for watching
Story 1- Human Selected Summary
Beijing unveiled official emblems are for the 2022
Winter Olympics and Paralympic Winter Games the
Winter Olympics emblem was inspired by the Chinese
character dong which means winter the upper half of
the logo was originated from the shape of a speed
skater while the lower part was from the skier the
emblem offer Winter Paralympics was transformed
from Chinese character Fei which means flying Real
Madrid will face a Brazilian club Jaime Oh in the World
Cup of football a final today in last six matches rial has
won four games while two matches ended inadrawon
the other hand out of their six matches Gremiohasa1-
lost - and two games ended in a draw in the semi-finals
Real Madrid beta al Jazeera - one while a cranial beater
patro up 1-0 in the semi-final match to reach the final
of the Club World Cup FIFA's ethics a watchdog
provisionally banned head of the Brazil's a Soccer
Association Marco Polo del Nero for 90 days the FIFA
ethics committee said that del Nero was banned from
all international and domestic soccer activities and
could be excluded for a further 45 days the ban was
imposed after a request from FIFA's the investigatory
champ chamber which is looking into unidentified
violations of the organization's ethics rules that's all in
this edition of fun news but before we go the Eiffel
Tower is all set to attract visitors during the Christmas
season with its transformation asawinterwonderland
complete visitors are being welcomed by a family of a
shiny penguins club chairs and replica of the Eiffel
Tower with mirrors we'll leave you with these are
stunning visuals thanks for watching In this above
example the extracted summary was highlighted in
yellow, and the ideal summary is highlighted in green.
From the above results, 8 sentences out of 15 were
matched by the generated summary. However, as
mentionedbefore,theidealsummarychosencannotbe
proven to be a perfect summary, therefore the match
ratio might not haveanymeaningtotheaccuracyofthe
system.
On the other hand, since the summary generated is an
extraction summary, i.e. extracts from original content
as summary, it is arguable that an accurate summary
can be provided by a few sentences of the original
content. The algorithm selects only 2 sentences from
each strong chain. Since there were no limitations on
the length of the sentences extracted, the accuracy of
the results varies.
Nevertheless, extracts from the original content will
have some indication to the topic of the news story,
which combined together with the key entities
detected, should serve as a reasonable summary for a
news story.
4.3 Keyword based News Retrieval System
Detected Key Entities (important keywords) are
proven to be most effective in indicating the topic of
the news story.WithidentifiedLocations,organization,
persons in the story, it greatly reduces the search time.
E.g. (To list all the news stories of ‘Narendra Modi’,
‘Cricket’ etc.).
The search GUI was designed to be a user friendly
interface, which reduces users’ learning time required
to adjust to the interface.
Figure-3: Keyword based News Retrieval System

5. CONCLUSION AND FUTURE WORK
5.1 Conclusion
The most of the people give primary preference to
broadcast news videos and the videos are regularly
seen by millions of people in the world. Developing a
system which generates extraction as well as
summarization for such News video and display
multimedia summary comprising video/audio and
texts as per user’s choice and interest is more efficient
and less time consuming.
In our project, we implemented Multimedia
summarization of news broadcastusinganchorperson
based story identification and Lexical chain algorithm.
Using this system one can generate multimedia
summary of one or more input news broadcast and
allow user to search and retrieve desired news story
using keyword based search and retrieval system.
Lexical Chain summarization was implemented to
provide summaries of news stories. Third party tools
such as WordNet, provided text recognition abilities
and vital sources of information which enabled the
implementation of such algorithms.
5.2 Future Work
 We can be extended our work to identify
anchor person automatically.
 This work can be extended to identify and
remove advertisement from summary.
 Improvement on Summarization
The quality of the news summaries have to be
improved. Investigationscouldbecarriedoutoneither
in searching for another generic text summarization
algorithm, or to improve the current algorithm. The
criteria in strong chain selection should be optimized
to extract more meaningful sentences.
REFERENCES
[1] Bailan Feng, Zhineng Chen, Rong Zheng, Bo Xu.
Multiple style exploration for story unit
segmentation of broadcast news video.
Multimedia System, Springer-Verlag Berlin
Heidelberg 2013
[2] Chaisorn, L., Chua, T.S., Lee,C.H.:Amulti-modal
approach to story segmentation for news
video. World Wide Web Internet Web Inf.
Systems 6(2), 187–208 (2003)
[3] Hemant Misra, Frank Hopfgartner, Anuj Goyal,
P. Punitha, and Joemon M. Jose,“TVNewsStory
Segmentation based on Semantic Coherence
and Content Similarity”, Dept. of Computing
Science, University of Glasgow, Glasgow, G12
8QQ, UK
[4] Jae-Gon Kim, Hyun Sung Chang, Kyeongok
Kang, Munchurl Kim, Jinwoong Kim, Hyung-
Myung Kim. “Summarization of News Video
and Its Description for Content-based Access”.
2004 Wiley Periodicals, Inc
[5] Kathleen McKeown and Dragomir R. Radev,
“Generating Summaries of Multiple News
Articles”, Department of Computer Science,
Columbia University, New York, NY 10027
[6] Kuan-Yu Chen, Berlin Chen and Ea-Ee Jan,
“Extractive Broadcast News Summarization
Leveraging Recurrent Neural Network
LanguageModelingTechniques”,T-ASL-04962-
2014.R1.R1
[7] Mark T. Maybury and Andrew E. Merlino,
“Multimedia Summaries of Broadcast News”,
Advanced Information Systems Center, the
MITRE Corporation, 202 Burlington Road
Bedford, MA 01730, USA
[8] Marcus J. Pickering, Stefan M. Rüger, “VIDEO
SEARCH ENGINE USING DUAL-MEDIA
SEGMENTATION”, Department of Computing;
Imperial College of Science, Technology and
Medicine; London
[9] Mark T. Maybury, “Discourse Cues for
Broadcast News Segmentation”, the MITRE
Corporation 202 Burlington Road Bedford,MA
01730, USA
[10] Peter Bell, Catherine Lai, Clare
Llewellyn, Alexandra Birch, Mark Sinclair.
“Asystem for automatic broadcast news
summarisation, geolocation and translation”,
Centre for Speech Technology Research,
University of Edinburgh, Edinburgh EH8 9AB,
UK
[11] Stokes, N., Carthy, J., Smeaton, A.:
SeLeCT: a lexical cohesion based news story
segmentation system. J. AI Commun. 17(1), 3–
12 (2004)
[12] Warren Greiff, Alex Morgan, Randall
Fish, Marc Richards, Amlan Kundu, “Fine-
Grained Hidden Markov Modeling for
Broadcast-News Story Segmentation”, MITRE

Corporation 202 Burlington Road Bedford,MA
01730-1420
[13] Wang, J.Q., Duan, L.Y., Liu, Q.S., Lu, H.Q.,
Jin, J.S.: A multimodal scheme for program
segmentation and representation in broadcast
video streams. IEEE Trans. Multimedia 10(3),
393–408 (2008)
[14] Xie, L., Zheng, L.L., Liu, Z.H., Zhang,Y.N.:
Laplacian ei- genmaps for automatic story
segmentation of broadcast news. IEEE Trans.
Audio Speech Language Process. 20(1), 276–
289 (2012)

IRJET- Segmenting, Multimedia Summarizing and Query based Retrieval of News Story from News Broadcast

More Related Content

Similar to IRJET- Segmenting, Multimedia Summarizing and Query based Retrieval of News Story from News Broadcast (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Segmenting, Multimedia Summarizing and Query based Retrieval of News Story from News Broadcast