Video Data Visualization System : Semantic Classification and Personalization

International Journal of Computer Graphics & Animation (IJCGA) Vol.2, No.2/3, July 2012
DOI : 10.5121/ijcga.2012.2201 1
VIDEO DATA VISUALIZATION SYSTEM: SEMANTIC
CLASSIFICATION AND PERSONALIZATION
Jamel SLIMI and Anis Ben AMMAR and Adel M. ALIMI
ReGIM (Research Group on Intelligent Machine)
National Engineering School of Sfax, Tunisia
National Engineering School of Sfax University of sfax, Tunisia
{jamel.slimi, anis.benammar, adel.alimi}@ieee.org
ABSTRACT
We present in this paper an intelligent video data visualization tool, based on semantic classification, for
retrieving and exploring a large scale corpus of videos. Our work is based on semantic classification
resulting from semantic analysis of video. The obtained classes will be projected in the visualization space.
The graph is represented by nodes and edges, the nodes are the keyframes of video documents and the
edges are the relation between documents and the classes of documents. Finally, we construct the user’s
profile, based on the interaction with the system, to render the system more adequate to its preferences.
KEYWORDS
Video, Semantic Concept Extraction, Semantic Classification, data Visualization, Personalization.
1. INTRODUCTION
Through the technological progress in video production and the availability of cheap digital
cameras, the size of video collections is growing tremendously in the personnel database or in the
database of dedicated Web sites such as “YouTube”, “FaceBook”, “DailyMotion”, etc [1]. Hence,
effective exploration and simple navigation in these collections become an urgent need.
Data visualization is still a relatively active field of research. Considering that vision is the most
developed sense in humans. It represents together with the brain psycho-visual vision and
cognitive capacities, a privileged tool to analyze one’s environment. Visualization research can
be defined by the techniques that allow transcribing or modelling complex phenomena or
structures, in order to manipulate and analyze them in visual forms.
Video data are very important sources of information. Owing to the fact that they cover most of
the daily events, they satisfy the extremely various requests of the public (political, cultural,
sport...) [2]. For these reasons, the video data are observed and analyzed by researchers.
Therefore, research fields focus more and more on video data processing such as video indexing,
concepts extraction, classification....
In recent years, there has been an interest in another type of challenge which is the exploration
and navigation in the video documents corpora. Despite the progress in video basic treatments,
automatic video data visualization is still suffering from the following problems.
The first problem resides on the semantic analysis of video data. For a good exploration and
navigation, visualization system must understand the basic semantics of the documents. Existing

2
video exploration systems can only support the services based on low-level visual features [3].
However using only low-level descriptors (form, texture, and colour) limits the ability of the
system. For this, we based our work on high level concepts to improve the quality of the video
data visualization system.
The second problem is how to represent the whole of the documents in the visualization space?
The visualization system is limited by the visualization surface. So, video collection (e.g.
thousands of hours of video) causes an organization problem of the keyframes in the visualization
space.
The third problem is how to personalize the visualization system behind receiving the user’s
input? Before the user’s interaction with the system to express his/her demand, the system can
only present a general overview of the corpus. This general overview can neither present the
details of the corpus nor response to the user preferences efficiently. For that, after the user input,
the system must reorganize the documents in an allowing way to disclose the details related to the
user’s input.
Based on the above elucidation, we propose a tool to help the user to explore a large corpus of
video document. The idea is to create a generic framework to fulfil various tasks in video data
visualization. Various tasks consist to the video concepts extraction, video data semantic
classification, video data organization and video data visualization system personalization.
The first step is devoted to the construction of the video data concepts vectors. The concepts
vector is composed of two objects which are: the keyframe that represents the document and the
whole concepts that describes it.
The second step is the semantic classification of the video data. The classification is resulting
from semantic similarity distance between concepts vectors describing video data.
The third step is the organization of the corpus through the projection of the classes in the
visualization space.
The last step is the creation of the user’s profiles via his/her interaction with the system to render
the video data organization more adequate to his/her preferences.
Compared to existing work, our contributions incorporate the following parts. First of all we treat
the various types of video documents. Unfortunately, the existing systems treats just one
type of video such as news, sports, etc. Secondly, we use high level concepts for the
description of the video sequences which has an important role in improving the semantic
classification. Thirdly, the visualization graph is represented in the form of a network based on
biological neuronal metaphor. This representation facilitates much navigation in the corpus of
documents. Finally and essentially, we integrate a personalization module in order to accelerate
the access process to the video documents. Personalization aims at representing the documents
according to the preferences and behaviour of the user.
This paper is organized as follows. Section 2 introduces a short state of the art on video
visualization system. Then, in section 3, we describe our framework. From sections 4 to 7 we
present the different tasks of the visualization processes. We, then, provide conclusions in Section
8 in which perspectives are given for future improvements.

3
2. PROBLEM POSITION AND RELATED WORKS
Due to the large amounts of multimedia data generates every day, an exploration and navigation
interface is becoming an increasing need. So data visualization represents a very active field of
research in the last years. Data visualization system attempts to exploit human visual processing
system innate ability to allow a user to perceive patterns, such as groups and outliers, within large
data sets in a more efficient manner than examination of the raw data.
The data visualization system interfaces permit the users to communicate with information in the
objective to finalize well defined tasks. The research tasks in interfaces field aim at defining
software methods, models and components for the creation of effective interfaces. Effectiveness
means the facility of training and use. Indeed, system interface plays a crucial role in the
development of interactive applications. Its ease of use can sometimes be a significant criterion in
the evaluation of the application. In addition, a good application which is represented by no
adequate interface can be judged as an unsuccessful application. In fact, the judgment of the
success of the data visualization interface resorts to the user expectations. So, interface must
satisfy the needs and the preferences of the user. However users do not have common perceptions
and choices. For that, the adaptability of the data visualization system interface becomes an
urgent need.
Various approaches exist for data visualization to help users explore in database and find
interesting documents intuitively. Proposed approaches are coupled with a variety of systems
such as multimedia retrieval engines [4] or annotation tools [5, 6].
Historically, visualization systems are interested in rearrangement of the research results in
textual database to facilitate navigation in the retrieved documents. Representations are based on
the similarity between the key words describing the documents [7]. Recently, there has been
several display systems treating the multi-media documents [8]. What follows is a non exhaustive
study of some systems followed by a comparative study. A comprehensive review on video data
visualization system can be found in [9].
Visualization systems based on timeline slider [10, 11, and 12] are the most common and easiest
way to get a quick overview of video content. The major disadvantages of these tools are the lack
of scalability for long documents and the unpleasant aspect due to the poor visual feedback.
Campanella et al. [5, 6, and 13] propose a data visualization system, FutureViewer, to explore and
annotate video sequences. The user can interactively visualize classes and annotate them and can
also graphically explore how the basic segments are distributed in the feature space. Each video
document is considered as a sequence of shots. The feature space is displayed in a 2D Cartesian
plane, where each axis corresponds to one feature type selected by the user and each shot is
represented by a little square filled by the dominant colour of the shot.
In the Intelligent Sensory Information Systems team of Amsterdam’s University, the MediaMill
video search engine [14, 15 and 16] proposes four browsing tools for video data. First, Galaxy
Browser uses 2D similarity-based visualization of keyframes where the user can annotate a large
collection of images. This system can improve the efficiency of the relevance feedback process
by mixing relevance feedback and 2D visualization [17, 18]. Second, Sphere Browser, represents
a novel interface for searching through semantic space using conceptual similarity. This is due the
classification shots with a similar conceptual index together into threads. The Sphere Browser
shows the time-line of the current video on the horizontal axis, and for each shot from the video it
displays the relevant threads on the vertical axis. Third, Cross Browser, uses a linear ordering to
ranking video data. The vertical axis is related to a selected concept. The horizontal one is used to

4
visualize video program in a time-line from which a keyframes is selected. The last, Rotor
browser, is an extension of the Cross Browser using more than two axes.
In [19], the proposed System, Table of Video Contents (TOVC), presents a framework for
multimedia content visualization and navigation. This tool is adapted for structured programs as
news, magazines or sport events. Connected with a player, users can have practical exploration
through a global view. The visualization shows the video structure and gives almost
instantaneously relevant information such as sequence number, relative durations, dominant
colours. . . .
In [20], the authors propose an “instant video browsing” which divides every video into as many
parts of equal length, as there are video windows opened on the screen. This tool offers two
different views for browsing the video data: the parallel view and the tree view. The limit of this
tool resides in the segmentation and the indexing of video data.
In [21, 22, and 23], the authors present initially a simple tool for storing images based on their
colour similarities. Just using the colour descriptor decrease the system performance. The
advantage of this tool is the organisation of the visualization space based on 3D cylinder form.
Secondly, they present a video browsing tool that supports users at the task of interactive search
and navigation. This tool is based in 3D user interface organised in the form of Thumbnail Ring
arranged by colour similarity.
Data clustering and intuitive representation are the two fundamental ideas developed over all the
proposed system. In fact, classification, whatever the inducted method or process, provides an
overview of all the data to be acceded. Visualization based on common and intuitive idea helps
the user to target the needed information. The major disadvantages of those tools are the
treatment of one type of video data (news or sport sequences) and essentially not having explored
the user’s attention models to enable more effective data visualization.
Our approach covers all the tasks of the visualization process: from indexing to the
organization of the video collection in the visualization space. The advantage of the
proposed approach consists to (1) the use of high-level features extraction (Visual
features, Text features, Audio features) for the description of the video data; (2) The
computation of the semantic similarity between the video data based on the correlation
between the concepts vectors describing the video collection; (3) the use of metaphor
inspired from the biological neuron function that allows a network representation in the
visualization space composed by nodes and edges; (4) the integration of a user model in
the visualization system allowing a representation adequate to the user preferences. In
result, our tool allows a simple, intuitive and efficient navigation in large-scale video
corpus.
In the following section we will detail our framework to represent and visualize large scale video
corpora.
3. FRAMEWORK DESCRIPTION
In this section, we present an overview of our plate form architecture. Our framework is
composed of four parts. The starting point of the data visualization system is a collection of video
data. This collection contains various type of video (film, sport, news…).

5
The first step consists of high-level concepts extraction which is based on video text extraction,
visual feature extraction and audio feature extraction. These concepts allow constructing the
concept vectors describing the video data.
The second step is dedicated to the construction of the similarity matrices. We calculate the
semantic similarity distance between all the concept vectors of video collection. The similarity
matrices are used to group elements and to explicit the relation between elements.
Figure1. Framework Description
The third step is devoted to the projection of the obtained classes (groups) on the visualization
space. The organisation of video collection is based on network representation. Key frames
represent the nodes of the representation and the edges represent the relations between elements.
The last step is dedicated to the personalization of the visualization system. The visualization
space does not make it possible to post the totality of the documents composing the corpus.
Personalization is expressed by the creation of the users’ profiles via their interaction with the
system. These users’ profiles will influence directly the visualization space and/or the semantic
distances calculation to post only the significant documents for the user.
The approach adopted for the realization of the visualization system has several advantages.
Firstly, we treat various types of video data. Also, the use of the high level concepts improves
semantic classification. Visualization in the form of a network explains the relations inter-
documents and facilitates the task of navigation in the corpus. The recourse to the user
preferences modelling allow a posting according to his/her needs that gives a fast and effective
access to the desired document.

6
4. INDEXING PROCESS
The video data collection used in our work, in the tests phase or for the development phase, is
extracted from TREC Video Retrieval Company (data base 2010). Several treatments are applied
to the data collection before being used. We based the multimodal indexing process on text
concepts extraction, visual feature extraction and audio feature extraction. The whole of the
extracted concepts are used to construct the concept vectors of video data. To start the concepts
extraction process video sequence must be divided on subsequence (shot). The shot consist of one
or more related frames that represent a continuous action in time and space [24]. For ease of use,
a shot is represented by a key frame. Generally, the key frame is the central frame of a shot [25].
Once, the video is segmented, we start the concept extraction process. When considering video,
the common approach is to take a camera shot as the pattern of interest. Such a pattern can then
be described using text features, audio features, visual features, and their combinations [26]
(Figure 2).
Figure2. Concepts Extraction
4.1. Video Text Extraction
A text embedded in video data contains large qualities of useful information. The first step of text
concept extraction consists to splitting the video data on frames. The second step consists of the
recognition of the embedded text in the frame. Video text extraction involves detection,
localization, enhancement and recognition of the textual content in the video frame [27]. To
segment the video data on frame, we have used the Java Media Framework package (JMF). The
input of this step is the shot containing text. At the rate of one frame per second, the shot is
framed into images. The obtained images are sent to an OCR for the text recognition.

7
Figure3. Video Text Extraction
4.2. Visual Feature Extraction
The aim of the visual feature extraction is to improve the video analysis. There are many visual
features [28, 29, 30, and 31]. We can divide visual descriptors into two main groups: General
information descriptors and Specific domain information descriptors. The general information
descriptor contains low level descriptors which give a description about colour, shape, regions,
textures and motion. The specific domain information descriptor gives information about objects
and events in the shot.
Colour: the key frame is an array of pixels (picture elements) where each pixel has a colour.
Colour histogram is the representation of the distribution of colours in the key frame [32]. HSV
space colour is defined as follow: Hue is the dominant wavelength in the color spectrum,
Saturation is a measure for the amount of white in the spectrum and Volume is a measure for the
brightness or intensity of the colour. Colour describing is based on the following tools: Dominant
Colour Descriptor (DCD), Scalable Colour Descriptor (SCD), Colour Structure Descriptor (CSD)
and Colour Layout Descriptor (CLD).
Texture: is an important point for keyframe describing. The texture descriptor observes the region
homogeneity and the histograms of these region borders. Descriptors are represented by:
Homogeneous Texture Descriptor (HTD), Texture Browsing Descriptor (TBD) and Edge
Histogram Descriptor (EHD).
Shape: Alternatively, a keyframe can be segmented by grouping pixels in the image based on a
homogeneity criterion on color, texture, or both [33, 34, and 35], or by connecting edge lines
[36]. Shape contains important semantic information due to human’s ability to recognize objects
through their shape. These descriptors describe regions, contours and shapes for 2D images. The
shape descriptors are the following ones: Region-based Shape Descriptor (RSD), Contour-based
Shape Descriptor (CSD) and 3-D Shape Descriptor (3-D SD).
Region: the Scale-Invariant Feature Transform (SIFT) descriptor measures the region around a
key point and describes each region using an edge orientation histogram. SIFT descriptor is the
best performing interest region descriptor. SIFT describe the local shape of the interest region
using edge histograms [37]. The interest region is divided into a 4x4 grid and every sector has its
own edge direction histogram. The grid is aligned with the dominant direction of the edges in the
interest region to make the descriptor rotation invariant.

8
4.3. Audio Feature Extraction
The text data derived from audio analysis is very important to describe video data. For
the audio feature extraction we use the Automatic Speech Recognition (ASR). The input
of the system is the video shot and the output is a text describing the speech included in
the video data. The use of the audio feature improves the video data description.
4.4. Supervised Learners
The Concept detection in video data is considered as a pattern recognition problem. For a pattern
x and a shot i, the aim is to obtain a probability measure indicating if the semantic concept Cx
exists in i. for this, we use the support vector machines (SVM) framework [38]. Also, we use the
LIBSVM implementation (for more detail sees http://www.csie.ntu.edu.tw) with radial function
and probabilistic output. The SVM yield a probability measure p(Cj|xi), which we use to rank and
to combine concept detection result.
5. SEMANTIC CLASSIFICATION
Semantic classification is the most important step in data visualization process. The aim of the
classification is to group video data according to their semantic concepts. These groups facilitate
the knowledge extraction task from the video data collection. Semantic classification is based on
video semantic Extraction. In this section, we present, firstly, the video data representation model,
after that we present our experiments about video data concept extraction and the concept
correlation, finally we present the similarity distance computation between the concept vectors
describing the set of video data composing the collection.
5.1. Video Data Representation Model
A video data is represented by two objects. The first is the keyframe and the second is the
concept vector which is composed by the high-level concepts extracted from video data. The
keyframe is the window representing the document in the visualization space. All the
modifications such as insertion, suppression, increase or decrease in size or the change of place
will be applied to this object. The keyframe is obtained by the segmentation of the video data.
For the document description, we use high level concepts. These concepts are extracted from text,
audio and visual constituting the video data. This is done a semantic description of the video.
Based on this semantic description we establish the phase of the similarity distances computation.
5.2. Concepts Extraction
The number of high-level concepts that was used in our work is 130 concepts (Figure 4). Each
document is characterized by a set of semantic concepts. These concepts will constitute the video
data concept vectors. The objective of the semantic concepts is to improve semantic
classification. Thus, we can facilitate the task of browsing the large scale corpora and accelerate
the access information task. The main objective of our system is to provide automatic content
analysis and an intelligent tool for retrieval and browsing a video data collection. Firstly, we
extract features from the keyframe and then label each keyframe based on corresponding features.
For example, if three features are used (colour, motion and audio), each keyframe has at least
three labels.

9
Figure 4. Concepts extraction
5.3. Concepts Correlation
The concepts used in the video data feature extraction process have a semantic relation between
them. The measure of this relation is based on correlation formula between concepts. The aim of
this correlation formula is to attribute a weight between a concept Ci and the other concept. This
weight is comprised between 0 and 1. 0 if there is not a relation between the two concepts and 1 if
C1 is a generalisation case of C2. For example the weight between “Adult” and “Person” is 1
because an adult is a person. But the weight between “Adult” and “Male_Person” is 0,5.
Figure three illustrate the result of the application of the following correlate formula.
B
and
A
indexing
shot
Number
A
indexing
shot
Number
Cd =
Figure5. Concepts correlation

10
5.4. Semantic Similarity distance
Semantic similarity or semantic relatedness is a concept whereby a set of documents or terms
within term lists are assigned a metric based on the likeness of their meaning /semantic content.
The determination of the degree of similarity between two semantic concepts is a difficult
problem which arises in many applications. The similarity calculation between concepts can be
based on the hierarchical link of specialization/generalization. The similarity between the
documents thus returns to a simple mathematical calculation between the vectors which compose
them.
We based the semantic similarity computation between two video shots on the correlation
formula highlighted in the precedent section. Let us consider two shots A and B. “Person”
and ‘Airplane” are the concepts describing A. “Actor” and “Building” are the concepts
describing B. The measure of the similarity between A and B is done by the summation
of the weight according to the concepts composing the descriptor vector of the shot A
(VA) devising by the number concepts constituting VA.
n
Vb
Va
Cd
B
A
Sd
n
∑
= 1
)
,
(
)
,
(
With Sd is the similarity distance, A and B are two different shots, Cd is the weight
between concepts describing A by report B and n is the number of concept composing
Vb.
The result of the semantic similarity distance is a semantic space, which is integrated in our work
(Figure 6). The semantic distance is comprised between 0 and 1. The distance is value 1 if the
two documents are similar and is value 0 if the two videos are completely different semantically.
If a user looks at two video documents which have 1 as measure semantic distance can note easily
that both videos indicate the same event. Both videos indicate the same event if they have the
same concept vectors.

11
Figure6. Semantic Space
6. VISUALIZATION SYSTEM
The main goal of the visualization space is to provide an overview of available data to user. A 2D
interface allowed visualizing the video data collection. Visualization is done by a projection of
the classes obtained after the similarity computation.
In literature, several data visualization system based their representation on the use of metaphor.
The use of the metaphor returns the interfaces more comprehensible and more meadows of
thought of the user.
In our approach, the metaphor is inspired from the biological neuron operation for representing
the collection in the visualization space. In the following section we explain the function of the
biological neuron.
6.1. Neuron metaphor
A neuron, also known as a neurone or nerve cell, is an electrically excitable cell that processes
and transmits information by electrical and chemical signalling. Chemical signalling occurs
via synapses, specialized connections with other cells. Neurons connect to each other to
form neural networks.
A typical neuron possesses a cell body (often called the soma), dendrites, and an axon. Dendrites
are thin structures that arise from the cell body. An axon is a special cellular extension that arises
from the cell body at a site called the axon hillock and travels for a distance. The cell body of a
neuron frequently gives rise to multiple dendrites, but never to more than one axon.
Hence, the body represents the node (keyframe), the axon represents the relation between the
classes and the dendrite represents the relation inter elements. In effects, the visualization graph is

12
in the form of neuronal network. The click on a keyframe is the stimulus which activates the
others keyframe related to them.
Figure7. Biological Neuron
6.2. Visualization graph
There is one conflicting requirements for visualizing large video data collection: The
visualization space is limited. So, we need to show as many elements of interest as possible in
such limited space. The network representation permits to decrease this problem.
The visualization graph is represented by keyframes and arcs. Arcs represent semantic relations
between documents and between classes. The keyframe is the object representing the documents
in the visualization space. Visualization in the form of network allows an effective browsing of
video data collection. This is done by explicit the semantic relation between the classes and
between the documents.
Visualization system posts a global overview of the corpus allowing an easy knowledge
extraction of the corpora. The problem is that the limited visualization space forbids the
revelation of the totality of knowledge. So the documents should well be chosen to be posted.
That always depends on the preferences and the expectations of users. The idea is to build user
profile allowing visualization on demand.
Figure8. Visualization Graph

13
7. PERSONALIZATION
Environment personalization consists in providing an appropriate environment to the user’s
profile. Profile is the regrouping of his/her preferences and his/her interest’s centers. The
personalization process is an incremental and interactive process in which the user is a decidor.
The creation of the user profile requires a collection of his preferences via its interaction with the
system. Before his input, the system posts a global overview of the video collection. This
overview reveals the possible maximum of topics existing in the video database.
7.1. User Interaction with the system
A user enters the system with an information need. The system posts an overview of the existing
topic in the video collection. The user chooses the more important shot by clicking in the
correspondent keyframe. This click is considered as stimulus which activates all the documents
with an important relation with the choosing element. This activation is according by posting the
interesting elements in the visualization space and eliminating the others. For example, when the
user clicks on a keyframe corresponding to the topic of sport, the system will give us only the
information’s that are related to this topic (for example animals). In the topic of sport, if the user
chooses football sequence, the documents which have an important relation with “football” must
be posted on the visualization space. For the reason that we cannot predict the following action
of the user, the system must keep a global overview of the corpus.
Figure9. Visualization space
7.2. User profile
The aim of user’s profile is to simplify the exploration of the video data collection. For this, the
knowledge interpretation must be easy to modify according to the user’s preferences. So, we must
weighting the edges connecting elements according to their importance for the user. Before, user
input System doesn’t have any idea about his preferences viewpoint. Hence, the construct of the
user’s profile needs the collection of his interest centers. This interest centers is done via the
interaction between the user and the system.
At anytime of the exploration, the user can express his interest viewpoint by clicking the node of
the network. Then, the system must modify the data representation by posting the most relevant

14
shot from the video collection. This by assigning a weight user on the relation between the
desired shot and the others shot composing the video collection.
)
,
( Si
Sd
Wu
R =
Where R is the relation between the desired shot Sd and the shot Si, Wu is the user weight. Wu is
comprised between 0 and 1.
The determination of the user’s preference is extracted from various indicators. The static
indicators describe the user personal information. The dynamic indicator which are extracted
from the interaction with the system: navigation history, frequency consultation and duration
consultation.
7. CONCLUSION AND FUTURE WORK
Visualization is an essential tool for exploring visual collections. To build a good visualization
system, a set of related requirements should be taken into account. In this paper we have
presented a video data visualization tool which covers the video data visualization process from
the video feature extraction to the personalization of the information access. The principal
objective of our tool is to simplify exploration, navigation and access to documents in large scale
video corpora. That through a network visualization graph (Nodes and arcs coloured).
The performance of the currently integrated visualization approach still needs to be improved.
The improvements will relate to the semantic classification part. Therefore, different semantic
classification techniques have to be studied. We plan to pass to a 3D representation of the
visualization graph. A long term goal of our work is to develop a generic video data retrieval
system.
REFERENCES
[1] Beecks .C, Skopal. T, Schoeffmann. K, & Seidl T,(2011) “Towards large-Scale Multimedia
Exploration”, 5th International Workshop on Ranking in Databases.
[2] Fan.J, Luo. H, & Elmagarmid. A. K,(2004) “Concept oriented indexing of video database toward
more effective retrieval and browsing”, IEEE Transactin on Image Processing, pp 974–992.
[3] AMARNATH. G, & RAMESH. J, (1997) “Visual information retrieval”, Communications of the
ACMMM .
[4] Snoek. C, Worring. M. Koelma. D, & Smeulders, (2006) “Learned lexicon-driven interactive video
retrieval”, In CIVR, pp 11.
[5] Campanella. M, Leonardi. R, & Migliorati, (2005), “Future viewer: an effcient framework for
navigating and classifying audiovisual documents”, In WIAMIS '05, Montreax, Switzerland.
[6] Campanella. M, Leonardi. R, & Migliorati, (2009), “Interactive visualization of video content and
associated description for semantic annotation ”, Journal In Signal, Image and Video Processing, Vol.
3, no. 2, pp. 183-196.
[7] Nguyen. G.P, & Worring. M, (2007), “Interactive access to large image collections using similarity
based visualization”, Journal of Visual Languages and Computing.
[8] Schoeffmann. K, Hopfgartner. F, Marques. O, Boeszoemenyi. L, & Jose. J.M, (2010), “Video
browsing interfaces and applications: a review”, SPIE reviews.
[9] Schaefer. G, (2011), “Image browsers Effective and efficient tools managing large image
collections”, IEEE International Conference on Multimedia Computing and Systems, pp 1-3.
[10] Hurst. W, Jarves. P, (2005), “Interactive, dynamic video browsing with the zoomslider interface”, in
IEEE ICME.

15
[11] Dragicevic. P, Ramos. G, Bibliowitcz. J, Nowrouzezahrai. D, Balakrishnan. R, & Singh. K, (2008),
“Video browsing by direct manipulation”, in SIGCHI on human factors in computing systems, pp
237-246.
[12] Shoeffmann. K, Boeszoermenyi, (2009), “Video browsing using interactive navigation summaries”,
In International Workshop on content-based Multimedia Indexing.
[13] Campanella. M, Leonardi. R, & Migliorati. P, (2005), “The future-viewer visual environment for
semantic characterization of video sequences”, In Proc. Of International Conference on Image
Processing, pp 1209-1212.
[14] Worring. M, Snoek. C.G.M, de Rooij. O, Nguyen. G.P, & Smeulders W.M, (2007). “The Mediamill
semantic video search engine”. In IEEE ICASSP, Honolulu, Hawaii, USA.
[15] Snoek. C.G.M, et all (2009), “The mediamill TRECVID 2009 semantic video search engine”. In
Proc. TRECVID Workshop.
[16] Snoek. C.G.M, Freiburg. B, Oomen. J, & Ordelman. R, (2010), “Crowdsourcing rock n’roll
multimedia retrieval”. In Proc. ACM Multimedia.
[17] Snoek. C.G.M, (2010), “The mediamill Search Engine Video”. In ACM Multimedia.
[18] Nguyen. G.P, & Worring. M, (2007), “Optimization of interactive visual similarity based search”, In
ACM TOMCCAP.
[19] Goeau. H, Thievre. J, Viaud,.M-L, & Pellerin. D, (2008), “Interactive visualization tool with graphic
table of video contents”. In CIVR.
[20] Del Fabro. M, Schoeffmann. K, & Boeszoermenyi. L, (2010), “Instant Video Browsing: A tool for
fast Nonsequential Hierarchical Video Browsing”. In Workshop of Intercative Multimedia
Applications.
[21] Schoeffmann. K, Taschwer. M, & Boeszoemenyi. L, (2010), “The video explorer: a tool for
navigation and searching within a single video based on fast content analysis”. In ACM Multimedia
Systems.
[22] Mûller. C, Smole. M, & Schoeffmann. K, (2012), “Demonstartion of a hierarchical Multi-layout 3D
video browser”. In IEEE Internatinal Conference on Multimedia and Expo.
[23] Schoeffmann. K, Ahlstrôm. D, & Bszormenyi. L, (2012), “Video browsing with a 3D thumbnail ring
arranged by color similarity”. In Advances in Multimedia Modeling, pp 639-641.
[24] Davenport. G, Smith. T.G.A, & Pincever. N, (1991), “Cinematic principles for multimedia”, IEEE
Computer Graphics & Applications, vol. 11, pp. 67–74,
[25] Brunelli. R, Mich. O, & Modena. C. M, (1999), “A survey on the automatic indexing of video data”,
Journal of Visual Communication and Image Representation, vol. 10, pp. 78–112.
[26] Snoek .C.G.M, Worring.M, (2008), “Concept-Based Video Retrieval”, Foundation and trend in
information retrieval, vol. 2, pp. 215–322.
[27] Karray .H and al (2008), “ReGIM at TRECVID 2008: high-level features extraction and video
search”, In TREC Video Retrieval Evaluation online proceedings.
[28] Datta. R, Joshi. D, Li. L, & Wang .J.Z, (2008), “Image retrieval: Ideas, influences and trends of the
new age”, ACM Computing Surveys, vol. 40, pp. 1–60.
[29] M. S. Lew, ed. (2001), “Principles of Visual Information Retrieval”. Springer.
[30] Smeulders. A.W.M, Worring. M, Santini. S, Gupta. A, & Jain. R,(2000), “Contentbased image
retrieval at the end of the early years”, IEEE Transactions on Pattern Analysis and Machine
Intelligence, vol. 22, pp. 1349–1380.
[31] Tuytelaars. T, & Mikolajczyk. K,(2008), “Local invariant feature detectors: A survey”, Foundations
and Trends in Computer Graphics and Vision, vol. 3, pp. 177–280.
[32] Van deSande. K, Gevers. T, & Snoek. C.G.M, (2010), “Evaluating Color Descriptors for Object and
Scene Recognition”, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 32.
[33] Carson. C, Belongie. S, Greenspan. H, & Malik. J,(2002), “Blobworld: Image segmentation using
expectation-maximization and its application to image querying”, IEEE Transactions on Pattern
Analysis and Machine Intelligence, vol. 24, pp. 1026–1038.
[34] Deng. Y, & Manjunath .B.S,(2001), “Unsupervised segmentation of color-texture regions in images
and video”, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 23, pp. 800–810.

16
[35] Hoang. M.A, Geusebroek. J.M, & Smeulders. A.W.M, (2005), “Color texture measurement and
segmentation”, Signal Processing, vol. 85, pp. 265–275.
[36] Gevers. T, (2002), “Adaptive image segmentation by combining photometric invariant region and
edge information”, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, pp. 848–
852.
[37] Van Germert .J.C, Geusedroek .J.M, Snoek .C.G.M, Worring .M & Smeulders .A.W.M, (2006), “the
challenge problem for automated detection of 101 semantic concepts in multimedia”, In proceeding of
the ACM International Conference on Multimedia, pp. 421–430.
[38] Byrne. D, Doherty. A.R, Snoek .C.G.M, (2010), “Everyday concept detection in visual lifelogs:
validation, relationships and trends”, In proceeding of Multimdia Tools Applications, pp 119–144.
Authors
Jamel SLIMI is a phd student in the National Engineering School of Sfax, Tunisia.
Member of the IEEE student Brench in Sfax, Tunisia. Member of the IEEE computer
society. Member of the ReGIM Laboratory (Research Groups on Intelligent
Machines).
Anis Ben AMMAR is assistant professor in Higher Institute of Business
Administration of Sfax, Tunisia. Member of the IEEE computer society. Member of
the ReGIM Laboratory (Research Groups on Intelligent Machines).
Adel M.ALIMI is professor in National Engineering School of Sfax, Tunisia.
Member of the IEEE computer society. President of the ReGIM Laboratory
(Research Groups on Intelligent Machines).

Video Data Visualization System : Semantic Classification and Personalization

More Related Content

What's hot (20)

Similar to Video Data Visualization System : Semantic Classification and Personalization (20)

Recently uploaded (20)

Video Data Visualization System : Semantic Classification and Personalization