Visual-CBIR: Platform for Storage and Effective Manipulation

of a Database Images

Kaouther Zekri

, Amel Grissa Touzi

and Noureddine Ellouze

ENIT, Tunis El Manar University, SITI, BP 37, Le Belvédère, 1002, Tunis El Manar, Tunis, Tunisia

ENIT, FST, Tunis El Manar University, LIPAH, BP 37, Le Belvédère, 1002, Tunis El Manar, Tunis, Tunisia

Keywords: Images Manipulation, Content-based Image Retrieval, Oracle DBMS, Search Queries, Signature, Similarity

Measure, Semantic Description.

Abstract: Today, image retrieval system has become a vital necessity for computing users. Different search systems

are increasingly invading the computing software markets, such as QBIC, Photobook and BlobWord. The

only negative point these systems have in common is their lack of semantics in query processing, low

interactivity and the irrelevance of search results. To overcome these limitations, we propose a more

efficient alternative system: a system for image retrieval. This new system provides an intelligent search by

content, by keyword, by index, etc. To confirm our approach, we have defined a combination with Oracle

DBMS that would lead to 1) an advanced modeling of image type using a signature that describes the

physical and semantic content of images, 2) the modeling of different types of search by creating stored

procedures in PL/SQL language and 3) simple storage and handling of images in database through an

intuitive interface. We prove that this system can be used in a distributed environment.

1 INTRODUCTION

Facing the speedy development of massive

computing information through the web, together

with technology progress, the issue related to data

management became more acute and important in

the context of information analysis and data

visualization. In this work, we focus on image data

that have experienced an evolution in terms of

storage, manipulation and content processing.

Any form of imagery that does not reflect a

specific field as content (science, geography,

physics, literature…) would hardly be classified

thus it would seem without any interest and would

end up in as a random result on a search list. This

kind of search results’ inexactitude proved the

necessity to develop Content-Based Image

Retrieval (CBIR) systems.

Since 1995, several search systems have been

commercialized, such as the systems presented by

Pecenovic (Pecenovic et al., 1998): IBM QBIC,

Photobook MIT Media Lab (Massachusetts

Institute of Technology), BlobWord of the

University of Berkeley in California, Virage and

Cortina. Despite their efficiency, these search

systems suffer from several limitations such as the

limited storage for database used relatively to fit

with the size of the current image database which

involves a different problematic with a very strong

constraints in time computing and the results

relevance. In 2005, Landré (Landré, 2005)

proposed a new architecture of CBIR based on the

hierarchical construction about increasing size

multiresolution signatures, the automatic grouping

of images in visually similar families of images

(database pre-classification), and the design of

images retrieval interface. This system suffers from

several problems such as non- trivial choice

parameters like the blur setting parameter during

the classification phase, which can lead to irrelevant

classification.

This paper aims to explain the system’s

procedure for image retrieval and to propose an

intelligent system for image retrieval (by content, by

keyword, by index, etc.) using Oracle DBMS. To

confirm our approach, we have defined a

combination with Oracle DBMS that would lead to

1) an advanced modeling image type using a

signature that describes the physical and semantic

content of images, 2) the modeling of different types

of search by creating stored procedures in PL/SQL

language and 3) a simple storage and handling of

images in database through an intuitive interface.

Zekri K., Grissa Touzi A. and Ellouze N..

Visual-CBIR: Platform for Storage and Effective Manipulation of a Database Images.

DOI: 10.5220/0005518900810090

In Proceedings of 4th International Conference on Data Management Technologies and Applications (DATA-2015), pages 81-90

ISBN: 978-989-758-103-8

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

The selection of Oracle DBMS is justified by

Oracle Multimedia (formerly Oracle interMedia)

feature that represents an extension to retrieve

multimedia data by their content. This extension

enables Oracle Database to manage, store, and

retrieve images using a new data type that stores the

inherent features of the image and using indexes to

make queries faster.

The second section of this work explains the

architecture of the image retrieval system and the

main query types that were used. In Section 3, we

discuss the reasons of our motivation. Our new

approach will be fully illustrated in Section 4.

Section 5 represents the validation and the

implementation of our system. Section 6 concludes.

2 IMAGE RETRIEVAL SYSTEM

2.1 Classical Architecture of

Content-based Image

Retrieval System

In a content-based retrieval approach, the descriptive

model associated with each image is represented as a

signature (or an image descriptor). This signature is

used to describe the visual content of the image. The

search is based on a comparison between a query

image and images database for classifying images

according to a distance measured between signature

of the query image and the images signature of the

database. Figure1 represents the classical

architecture of CBIR systems.

Figure 1: Classical architecture of a CBIR system.

The architecture of this system is divided into

two phases of treatment:

 The Offline Phase: has a characterization

step in which attributes are automatically

extracted from the images of the database and

stored in vectors. Then, these vectors are

stored in a database of descriptors.

 The Online Phase: aims to extract the

descriptor of the query image given by the

user and compare it with descriptors in

database to select similar images to the query

image.

The system returns the search result in a list of

images ordered according to “the similarity

measure” between their signatures and the signature

of the query image.

2.2 Example of Existing CBIR Systems

The development of several search systems allows

us to facilitate the CBIR task. Among these systems

we can provide: QBIC (Query By Image Content)

(Flickner et al., 1995) of IBM, one of the first search

systems which deals with the image search based on

color, texture, shape and sketch. The BlobWorld

(Carson et al., 1999) of the University of California;

Berkeley; works on homogeneous areas from the

image. It allows recovering from a motion picture,

similar regions in color and texture. Photobook

(Piccard et al., 1996), developed by MIT Media Lab,

offers a powerful search to collect homogeneous

images. The search is possible on three different

criteria: appearance, shape and texture. For recent

systems, we can find img(Rummager) and the

website img(Anaktisi) (Zagoris et al., 2009. ) which

can execute an image search based on query image

with a new set of descriptors which include the

characteristic information of the color and texture in

histogram. Despite their efficiency, these approaches

suffer from several limitations, such as: small image

databases used the lack of semantics in the query

processing, imprecision of results and the lack of

integration of image processing techniques.

The MedFMI-SiR system (Daniel et al., 2011)

represents an open source module that executes

similarity queries over DICOM images integrating

metadata- and content-based image retrieval. This

solution is attached onto the Oracle DBMS,

benefiting from all the capabilities it natively

provides. The main drawback of MedFMI-SiR is

that it uses a specialized method to perform only a

feature extraction of the DICOM type image.

In his approach, Landré proposed a CBIR

system using MySQL DBMS. He represents search

techniques for images using visual navigation in the

database. His system organizes descriptors

according to an increasing size to classify images

into families based on multiresolution signatures.

However, this system suffers from the irrelevant

classification in case of non-trivial choice o f

parameters like the blur setting parameter.

DATA2015-4thInternationalConferenceonDataManagementTechnologiesandApplications

Moreover, this approach does not support an

integrated search query which joins textual

information and visual content in the retrieval

process.

2.3 Image Retrieval System and DBMS

Faced with a great quantitative evolution of images,

the current DBMS developed their storage and data

management capabilities in order to support the

image databases. As an example of DBMS, we

quote: MySQL and Oracle (Gabillaud, 2009).

In (Landré, 2005), Landré proposed a system of

image retrieval based on the MySQL DBMS. The

architecture of this system consists of a web server

(Apache) that manages dynamic web pages (PHP)

able to access data from the relational database

(MySQL).

In this search system, the construction of the

signature involves the following steps: the wavelet

transformation of images, the extraction of

multiresolution attributes form descriptor vectors,

the organization of attributes in signature vectors

stored in MySQL database and the classification of

signature vectors in families according to distance.

The CBIR task and the building of the image

signature are detached from the MySQL DBMS and

executed in a separate retrieval engine. Only storage

and metadata-based queries are realized using the

MySQL database.

Oracle database is a relational database

management system (RDBMS) that has introduced

multimedia applications and object-oriented

development since version 8. Moving to Oracle

Database 11g, Oracle Multimedia (previously

interMedia) becomes a feature that extends Oracle

Database reliability, availability, and data

management to multimedia content in traditional,

Internet, electronic commerce, and media-rich

applications (Rod et al., 2001).

The Oracle Multimedia architecture defines the

framework through which media content as well as

traditional data are supported in the database (Rod et

al., 2001). This architecture includes three principal

levels: Oracle database server, Oracle database, and

client application. Oracle database server is the

software that manages Oracle databases, and client

applications that interact with the server to add,

update, retrieve or delete data.

Using Oracle multimedia feature, Oracle

Database can store, manage, retrieve, and

manipulate multimedia data, especially the image

data. Oracle supports CBIR task that allows image

retrieval using the Oracle Multimedia OrdImage

object type. This can give the possibility to relieve

image database management and to build new end-

user image retrieval applications. In Oracle

Multimedia, the CBIR has been adopted as a

complementary technique to Text-Based Image

Retrieval (TBIR) (Li et al., 2011).

The type ORDImage provides the ability to

manage all image information as attributes (also

called image metadata). These attributes include:

source for the storage information, contentLength,

height and width of the image, fileFormat,

compressionFormat, contentFormat, and mimeType.

This image metadata is collected and organized in

schema-based XML documents. The metadata are

useful to help search in image datasets by indexing

the metadata for powerful text and thematic media

retrievals using

Oracle Text (Rod et al., 2001).

Oracle Text provides a linguistic analysis on

documents, as well as retrieval text using a variety

of strategies performing keyword searching, context

queries, pattern matching, HTML/XML section

searching, Boolean operations, mixed thematic

queries, and so on.

Oracle Multimedia is a single integrated feature

that enables to integrate the qualities of the CBIR

systems and the TBIR systems into a unique

application (Dimitrovski et al., 2009.). Combining

content-based and text-based image retrieval in an

integrated retrieval engine can lead to a better

accuracy, because one complements the other.

3 MOTIVATION

The approaches proposed in different search systems

such as QBIC, Photobook, Virage and Cortina

represent fairly important work in the Content Based

image Retrieval area. However, these search systems

suffer from several limitations such as the limited

storage for database, the difficulty of handling the

stored images and the image retrieval task with very

strong constraints in time computing, and the

irrelevance of search results.

The approach proposed by Landré represents an

image retrieval system based on multiresolution

signatures stored in MySQL database. For the

signature construction, as we have seen in the

previous section, Landré was forced to go through

several steps to build image signatures in MySQL

DBMS, which does not support the content based

image retrieval task. In this approach, a limitation

comes from the non-trivial choice of parameters

such as blur setting parameter during the

classification phase, which can lead to irrelevant

Visual-CBIR:PlatformforStorageandEffectiveManipulationofaDatabaseImages

data classification. Moreover, this system doesn’t

support the integrated search query which combines

textual-based and content-based image retrieval.

Oracle Multimedia CBIR shows very good

results of image search compared to other active

retrieval systems tested by the proposed

benchmarking system for CBIR in (Harald and Paul,

2010). Moreover, it provides the means necessary to

build an integrated retrieval engine that can execute

optimized search query by combining textual- and

content-based image retrieval.

In section 4, we will propose our new CBIR

approach based on Oracle DBMS.

4 NEW APPROACH

4.1 Principle of Our Approach

In this section, we represent a new approach of

CBIR system based on Oracle database. Our

approach is inspired from a system proposed by

Landré, MedFMI-SiR system, and Oracle functions

to manage the content of the images. To overcome

the difficulties of retrieval and manipulation of

image data in the retrieval systems, such as we have

described in the previous section, we proposed a

new content-based image retrieval system that

allows 1) an advanced modeling of image type using

a signature that describes the physical and semantic

content of images, 2) the modeling of different types

of search by creating stored procedures in PL/SQL

language, 3) a simple storage and handling of

images in database through an intuitive interface,

and 4) an integrated search query combining text-

based and content-based image retrieval.

In the construction phase of feature signatures,

we can distinguish between two types of features:

physical and semantic features (Atnafu Besufekad,

2003).

 Physical Features: or low-level features that

describe the visual content of image. For

example: “how closely its color and shape

match a picture of a specific object”.

 Semantic Features: or high-level features

that describe the semantic content of the

image: they describe the nature of the

relationships that link objects within an image.

We can use traditional text to describe the

semantic significance of the image. Example

of a query type for semantic content: “images

of tumor in the right lobe of the brain”.

In our system, the user disposes of a sample

image to make his query. The search queries relate

to color, texture, shape, location of these criteria in

the image, and a combination of these different

search modes. The system allows you to assign a

weight to give more or less importance to certain

search criteria, such as shape, texture, color and

location. Thus each criterion is assigned a

coefficient representing the importance attached to

it during the comparison.

For text-based image retrieval task, the system

allows the user to assign a textual description to one

or a set of visually similar images when they are

added to the database. After storing the images, the

system extracts all image

content and a rich set of

metadata attributes. Once the image metadata has

been extracted and stored, the system indexes the

metadata and keywords assigned for each image for

powerful thematic image retrieval based on Oracle

Text. Thus, the database can locate easily the image

data with indexes that can speed up queries based on

metadata and keywords stored with the image.

Moreover, our system can allow users to provide both

textual- and content-based image retrieval to define

the desired result.

The interface of our system provides the ability

to visualize the images in the database in order to

have a clear idea of what we want and to manage

the contents of the database by adding or removing

images.

As has already been explained, we will treat the

architectures proposed by Landré and MedFMI-SiR

system by using Oracle DBMS to develop our

CBIR system. In addition, we will try to extract

physical features (visual content) and semantic

features (textual information) from images in order

to improve and speed up our search by indexing the

signature and the textual information of image.

4.2 Architecture of Our Approach

The architecture of our system is shown in Figure 2.

The system provides an interface for easy

management of image by storing, removing images

and visual allowing navigation in database. Our

approach replaces the construction sequence of

image signatures (feature extraction, descriptor

vectors, and signature vectors) by a single step of

generating signatures using Oracle functionalities.

At the end of each step of retrieval phase, an option

to generate a SQL script for each data manipulation

will be given to visualize the execution validity of

tasks in complete transparency.

DATA2015-4thInternationalConferenceonDataManagementTechnologiesandApplications

Figure 2: Architecture of our CBIR system.

The content image retrieval process in our

system is based on extracting color, texture, shape,

and where in the image these features are locating

and comparing them. The features are extracted by

1) segmenting the image into regions according to

color, 2) determining the features for each region.

Color and texture information are determined

globally, by unifying the region-based information,

to generate global color and texture histograms. The

region, texture, global color, and shape information are

stored in the signature to represent these attributes for

the entire image

. This generated signature is stored in

the signature-database.

For the online phase, when the user selects a query

image, the system is responsible for registering the

image in a temporary table in the Oracle database to be

able to generate its signature. Then, a comparison will

be made automatically with the database signatures to

return a list of selected images whose content is very

close to the query image. To measure the distance

between signatures in comparison phase, the system

uses the Manhattan distance.

To execute integrated image retrieval, the system

allows posing integrated queries for text-based and

content-based search. After selecting an example

image, the user enters a few keywords that describe the

content of the image he seeks. The textual description

and the generated signature of the example image will

be used to locate and select from the indexed database

the images that share the same visual features and

textual information with the proposed query.

Finally, we can deal with the phase of

interpretation of the result obtained by the textual

description of the semantic content of the selected

images. This last step is based on the keywords

extracted from database for all selected images in

the search phase in order to semantically describe

the content of images.

5 VALIDATION OF

CONTENT-BASED IMAGE

RETRIEVAL BASED ON

ORACLE DBMS

5.1 Implementation of CBIR System

and Database under Oracle

The implementation of a CBIR system in Oracle is

always a difficult task especially with the huge mass

and different types of images stored in database. In

fact, this implementation, hand-signed by the

designer, can make the CBIR like a delicate task that

requires a thorough knowledge of Oracle database

with a remarkable lack of transparency in handling

image data.

Usually, the designer should stick to the

following measures:

 Creating the images database.

 Textual description of the database content.

 Manual selection of search criteria

corresponding to the type of images database.

 Defining PL / SQL scripts for each operation

performed in the database to make image

management and search task automatic and

transparent.

The designer must define a threshold of

similarity measurement between images in database

in order to select the list of closest images to the

image query.

To illustrate the tasks required to implement our

CBIR example, we represent in the following a set

of scripts to implement two tables images and

image_signature, add an image (Figure 3) to the

database and implement the CBIR task. For

example, the following script can be issued to create

images, image_metadata, and image_signature

tables to store the image with its metadata and

signature:

-- images table

CREATE TABLE images (

id INTEGER PRIMARY KEY,

image ORDSYS.ORDImage,

description varchar2(255));

-- image_metadata table

CREATE TABLE image_metadata(id

INTEGER PRIMARY KEY REFERENCES

images(id),

metaORDImage XMLTYPE,

XMLType COLUMN metaORDImage

XMLSCHEMA http://xmlns.oracle.com

/ord/meta/ordimage

ELEMENT "ordImageAttributes"

Visual-CBIR:PlatformforStorageandEffectiveManipulationofaDatabaseImages

-- image_signature table

CREATE TABLE

image_signature (

id INTEGER PRIMARY KEY

REFERENCES images(id,

signature_img

ORDSYS.ORDImageSignature);

where OrdImage is the Oracle’s abstract data type to

store images. If the image content is stored in the

database, it can be handled as another relational data

using SQL. The ORDImageSignature object type

supports content-based image retrieval, or image

matching, where signature_img is an attribute that

stores image signature. metaORDImage is the

column where the ordimage metadata will be stored.

It is bound to the XML schema stored at

http://xmlns.oracle.com/ord/meta/ordimage, and

defined as the XML element ordImageAttributes.

To add an image to Oracle database, we propose

an example of a non-expert user of images database

who wants to insert a "cerebrale.jpg" (Figure 3) file.

This image is situated in a logical folder

"IMG_DOSSIER" created under oracle by the user.

Figure 3: A medical image selected to be added to the

database.

By adding an ORDImage to the database we

intend to the addition of an image and the generation

of its signature in an ORDImageSignature type

attribute. The set of visual attributes is stored in a

BLOB. Before the generation of an image signature,

it must, like the type ORDImage call a builder, called

init(), which is used to initialize the BLOB attribute

to an empty BLOB. Once this realized, we can call

the method generateSignature by giving the

parameter ORDImage attribute to obtain its visual

characteristics.

--initialization and image insertion

INSERT INTO images (id, image)

VALUES(id_val, ORDImage.init());

INSERT INTO image_signature (id,

signature_img) VALUES (id_val,

ORDImageSignature.init());

SELECT i.image, s.signature_img into

obj, obj_sign from images i,

image_signature s where i.id= s.id

and s.id=id for update;

obj.setSource('FILE','IMG_DOSSIER',

'cerebrale.jpg');obj.import(ctx);

UPDATE images set image = obj where

id=1;

--generation of image signature

obj_sign.generateSignature(obj);

UPDATE image_signature p SET

p.signature_img = obj_sign WHERE

id=id_val;

COMMIT;

END;/

To perform the CBIR task, the user needs to

implement the following PL/SQL script:

DECLARE

img_sig ORDSYS.ORDImageSignature;

img_sig_req

ORDSYS.ORDImageSignature;

-- Specify a weight for criteria

commande varchar(200) := 'color=0

texture=1 shape=0 location=0';

eval_score float;

threshold float;

similaire int;

BEGIN

SELECT p.signature_req INTO

img_sig_req FROM image_requete p

WHERE p.id_req = 1 FOR UPDATE;

FOR record_img IN (SELECT id,

signature_img FROM

image_signature

FOR UPDATE) LOOP

-- Comparison between signatures

threshold := 20;

similaire :=

OrdImageSignature.isSimilar

(record_img. signature_img,

img_sig_req, commande,

threshold);

DBMS_OUTPUT.PUT_LINE('

isSimilar'|| similaire);

END LOOP;

COMMIT;

END;/

In the retrieval phase, it is necessary to make the

comparison between the signature of an image query

and all image signatures stored in the database. The

image signature is obtained through an Oracle

analysis of the image with generateSignature()

method of ORDImage type. This signature contains

color information, textures and shapes according to

each area of the image. It also contains information

on the image background. The result of this analysis

is then contained in a type called

ORDImageSignature.

DATA2015-4thInternationalConferenceonDataManagementTechnologiesandApplications

Seen that the database is a medical images

database, the user must give more importance to the

texture criterion to execute a texture-based retrieval

query. Therefore, the user must assign a weight in

the range 0 to 1 for color, shape, texture, and

location. To specify a high importance to the texture

criterion, the user should select a high weight for

texture. In our application, a high weight represents

a value of 1, a medium weight is a value of 0.5, and

a low weight represents a value of 0.

The operator used for ORDImage object type is:

isSimilar(). This operator compares the signature of a

query image with the signatures of images stored in a

database, and determines whether or not the images

match, based on the weights assigned to the search

criteria and threshold value. The comparison of two

signatures returns a score between 0.0 and 100.0,

where a lower value indicates a closer match. The

choice of threshold is empirical and it is vital to try to

model and control the choice of appropriate threshold

for a specific image type. Currently, we use an

approximate value threshold that does not exceed 20.

The value of the threshold must be comprised

between 0 and 100, which is the range of the

distance. Then only images whose signatures are a

distance (score) of 20 or less from the query

signature will be selected.

To improve the search performance and easily

execute the task of CBIR, we have implemented an

index-based retrieval on the image signature. This

index is of type ORDImageIndex. By defining this

index item we can filter and locate specific

information and then retrieve it as efficiently as

possible. Therefore, we can have an advantage of the

increased performance based on image matching

with image signature indexes on the signatures

database. The following command creates an index

on image_signature table, based on the data in

the signature_img column:

CREATE INDEX indexSign ON

image_signature

(signature_img)

INDEXTYPE IS ORDSYS.ORDIMAGEINDEX

PARAMETERS

('ORDImage_Filter_Tablespace =

<name>,ORDImage_Index_Tablespace =

<name>');

To easily locate the image data based on textual

information, we implemented a full text index based

on Oracle Text. Oracle Text is an extensive full text

indexing technology that uses standard SQL to

index, analyze, and search text and documents stored

in the Oracle database. It can be used to index XML

data. All image metadata is extracted in the form of

XML database objects and returned as a database

XMLSequence type. This XML format is easier to

be used with database features and to be searched

with Oracle text. The following command represents

a query that selects the OrdImage attribute of image

compressionFormat:

SELECT i.image

FROM images i, image_metadata e

WHERE i.id = e.id

AND extractValue(metaordimage,

'/ordImageAttributes/

compressionFormat/text()',

'xmlns="http://xmlns.oracle.com

/ord/meta/ordimage"')= 'JPEG' ;

where extractValue is the function that scans the

XML data and identifies the elements that satisfy the

given query. Image retrieval based on Metadata

queries can be efficiently performed using a b-tree

index which can be created on the extractValue

function result. This index is employed to

dramatically increase the speed of the previous

query:

CREATE INDEX indexmeta

ON images e

(extractValue(metaordimage,

'/ordImageAttributes/

compressionFormat/text()','xmlns="

http://xmlns.oracle.com/ord/meta/

ordimage"'));

For the text retrieval, we used the context index

type of Oracle text indexes. In the following

command, we represent an example of context index

created on the description column that contents

semantic descriptions assigned for each image stored

in database.

CREATE INDEX textindex

on images (description)indextype

is ctxsys.context;

The integrated search allows posing queries

mixing the textual and visual information to achieve

a better accuracy and flexibility regarding query

resulting. By using a combination of different search

queries which we have already executed, we were

able to implement an integrated query which allows

the selection of images whose signatures are the most

similar to the given example image, and have a

semantic description about brain with tumor. This

query selects only the image which characterized by

JPEG compression format:

SELECT i.image FROM images i,

image_signature s, image_metadata m

WHERE i.id = s.id AND i.id=m.id

AND ORDSYS.IMGSimilar

(s.signature_img, signature_query,

Commande, threshold)=1

Visual-CBIR:PlatformforStorageandEffectiveManipulationofaDatabaseImages

AND contains( description,

'brain,tumor') > 0

AND extractValue (metaordimage,

'/ordImageAttributes/

comprssionFormat/text()','xmlns=

"http://xmlns.oracle.com/ord/meta/

ordimage"')= 'JPEG';

Therefore, our system makes possible to integrate

content-based and text-based retrieval for efficient

image search by

executing queries through indexed

searching over triple image similarities, textual

descriptions, and metadata attributes.

5.2 Intelligent-CBIR System

To remedy the lack of transparency and the

implementation sensitivity of Oracle CBIRS, we

propose a new layer that allows efficient

manipulation of images data type and search by

content in full transparency through a simple and

intuitive interface. To implement our tool, we used

Microsoft Windows Seven software environment.

We use the Oracle 11g release2 database. For the

development environment we use NetBeans IDE

7.4.

Intelligent-CBIR system provides designers with

multiple screens. We illustrate some examples as

follows:

Once the user chooses to connect to the database

by clicking on the corresponding visual navigation

button, the system loads the images in the database

to allow the user to have an idea about the database

content (Figure 4).

Using visual navigation, the user can get a

general idea about the database content that will help

him easily manipulate images by adding or deleting

options. Figure 5 shows an example of adding an

image to the database. To add an image to the

database, you simply select an image through the

system interface and validate your choices. SQL

script will be displayed for each handling contents of

the database for further data manipulation

transparency in our system.

For the task of CBIR, the system allows selecting

an image query and selecting search criteria with

their degree of importance according to the

information provided by the system or the user's

choice as shown in Figure 6:

 The Color Criterion: shows the distribution

of colors in the entire image.

 The Texture Criterion: represents the low-

level models and textures in the image.

 The Shape Criterion: represents the forms

that appear in the image. They can be

determined by techniques based on color

segmentation.

 The Location Criterion: shows the positions

of shape, color and texture.

Figure 4: Interactive visual browsing of images databases.

Once the user validates the choice of search

criteria and starts the images retrieval, the system

browse through all images in database and compares

the signature of the image query by those of images

database.

Figure 5: Adding an image to the database.

For each comparison, a value of similarity

measurement is associated with each image: all

images, whose value of similarity measurement is

lower than a threshold (specified by the

administrator), will be selected and displayed as

search results (Figure 6).

5.3 Type of Search Query in Our

System

This solution offers the user the flexibility for

handling the choice of search criteria, using a single

criterion or a combination of criteria, which allows

DATA2015-4thInternationalConferenceonDataManagementTechnologiesandApplications

multiple types of queries such as: query by color,

query by texture, and query by shape and texture etc.

 Query by Color: search request relates to the

color feature that is used to measure the

similarity between the images based on the

color criterion. This query gives more

importance to color in the comparison phase

between images.

 Query by Texture: search request relates to

the texture feature that represents the visual

patterns having homogeneity properties

which do not result from the presence of a

single intensity or color only. Texture

determination is ideally suited for medical

image retrievals (Lehmann, 2005).

 Query by Shape: search request relates to

the shape feature to measure the similarity

between different shapes contained in the

images.

 Query by Combination of Different Search

Criteria: search query involves a

combination of criteria. Example of this type

of query "homogeneous texture images and

even distribution form."

 Query by Keywords/Metadata: search

query based on a set of keywords which can

describe the semantic content of the image.

The metadata-based query performs an image

retrieval based on the attributes attached to

each image file which can be extracted and

stored in the database.

 Integrated Query: this type of query is used

in order to optimize the search task. We have

achieved a combination of textual search (for

the semantic content of the image) and

content search (for the physical content of the

image) in order to enhance the results and

provide a semantic interpretation of the

image content.

Figure 6 illustrates an example of image

retrieval based on the texture and shape criteria.

Since we are working on a database test containing

medical images, we chose to give more importance

to the texture criterion (Thomas, 2004) and less

importance to the color, shape and location criteria.

After each comparison, the system displays a

measure of similarity (score) for each image, which

is the rate of correspondence between the query

image and the images of our database. Only images

which have a score below a certain threshold will

be selected.

Figure 6: Content-based image retrieval interface.

5.4 Comparison with -- Other

Approaches

Table 1 shows a comparison between the

functionality provided by our system, the Landré

approach and MedFMI-SiR system.

Table 1: Comparative table between our CBIR system and

existing systems.

Functionalities

Landré

approach

MedFMI-

SiR

Our

system

Multi-types of

content search

Search by physical

content

Search by

semantic content

Semantic

interpretation

Manipulation of

images database

Integrated search

queries

In our system, we were able to model the image

data by the feature extraction and semantic

interpretation of the visual content of the image.

We make possible to integrate text- and content-

based retrieval in order to execute an efficient

image search. We could also provide a simple and

transparent handling (compared to other systems) of

data stored in the Oracle database by adding or

deleting images through a simple interface of our

system.

Visual-CBIR:PlatformforStorageandEffectiveManipulationofaDatabaseImages

6 CONCLUSIONS

With the great mass of data stored in image

databases, content-based image retrieval has become

a necessity in order to classify images and extract

useful information from this large amount of data. In

this approach we have developed a simple and

intuitive interface which ensures an advanced

manipulation of images using Oracle database. One

of the advantages of using a DBMS to manipulate

images is to be able to search for images in many

ways, as well as using a centralized manageable

repository.

Through the study of the implementation manner

of the content based image retrieval in Oracle and

taking into consideration the absence of a simple and

intuitive interface that allows user to do an

intelligent and automatic search for images in

database, we decided to create a layer of assistance

to design and implement CBIR system. The result of

this work is a search system that allows visual

navigation, manipulation of the image database and

CBIR. In addition, Oracle is a distributed DBMS

(Özsu and Valduriez, 2011); so we can model our

system directly into a distributed environment.

The CBIR in our approach is based only on the

Oracle provided features. The future proceedings

also involve integration of our own feature

extraction and signature construction methods.

REFERENCES

Atnafu Besufekad, S., 2003. Modélisation et traitement de

requêtes images complexes. Doctoral dissertation,

National Institute of Applied Sciences of Lyon, Lyon,

216 pages.

Carson, C., Thomas, M., Belongie, S., et al., 1999.

Blobworld: A System for Region-Based Image

Indexing and Retrieval. In Visual Information and

Information Systems, Third International Conference,

VISUAL'99, volume 1614, pages 509–517. Springer.

Daniel, S. Kaster, Pedro, H. Bugatti, Marcelo Ponciano-

Silva, et al., 2011. MedFMI-SiR: A Powerful DBMS

Solution for Large-Scale Medical Image Retrieval. In

Information Technology in Bio- and Medical

Informatics, Second International Conference, ITBAM

2011, volume 6865, pages 16–30. Springer.

Dimitrovski, I., Guguljanov, P., and Loskovska, S., 2009.

Implementation of web-based medical image retrieval

system in Oracle. In Adaptive Science & Technology,

2009. ICAST 2009. 2nd International Conference on,

pages 192–197. IEEE.

Flickner, M., Sawhney, H., Niblack, W. et al., 1995.

Query by image and video content: The qbic system.

In Computer, 28(9), 23–32. IEEE.

Gabillaud, J., 2009. Oracle 11g: SQL, PL/SQL, SQL*Plus.

France: Editions ENI, 483 pages.

Harald, K., and Paul, M., 2010. Content-Based Image

Retrieval Systems - Reviewing and Benchmarking. In

Journal of Digital Information Management, volume

8, pages 54–64.

Landré, J., 2005. Analyse multirésolution pour la

recherche et l’indexation d’images par le contenu

dans les bases de données images – Application à la

bases d’images paléontologique Trans’Tyfipal.

Doctoral dissertation, University of Burgundy -

France, 159 pages.

Lehmann T. M., Güld, M.O., Deselaers, T., et al, 2005.

Automatic categorization of medical images for

content-based retrieval and data mining. In

Computerized Medical Imaging and Graphics,

Elsevier, 29(2-3), pages 143–155.

Li, W., Duan, L., Xu, D., and Tsang, I.W., 2011. Text-

based image retrieval using progressive multi-instance

learning. In Computer Vision (ICCV), 2011 IEEE

International Conference on, pages 2049–2055. IEEE.

Özsu, M. T. and Valduriez, P., 2011. Principles of

Distributed Database Systems, Springer-Verlag New

York. Third Edition, 845 pages.

Pecenovic, Z., Do, M., Ayer, S., and Vetterli, M., 1998.

New methods for image retrieval. In Proceedings of

the International Congress on Imaging Science,

volume 2, pages 242–246.

Piccard, R. W., Pentland, A., and Sclaroff, S., 1996.

Photobook: Content-based manipulation of image

databases. In International Journal of Computer

Vision, volume 18, pp. 233–254. Springer.

Rod, w., et. al., 2001. Oracle interMedia User's Guide and

Reference, Release 9.0.1, Part No. A88786-01, [on

line], (30 March 2014) http://docs.oracle.com/html/

A88786_01/title.htm.

Zagoris, K., Chatzichristofis, S. A., Papamarkos, N., and

Boutalis, Y.S., 2009. Img(Anaktisi): A Web Content

Based Image Retrieval System. In SISAP '09

Proceedings of the 2009 Second International

Workshop on Similarity Search and Application, pages

154–155. IEEE.

DATA2015-4thInternationalConferenceonDataManagementTechnologiesandApplications