Content uploaded by Nilanjan Dey
Author content
All content in this area was uploaded by Nilanjan Dey on Jun 25, 2015
Content may be subject to copyright.
Int. J. Image Mining, Vol. 1, No. 1, 2015 45
Copyright © 2015 Inderscience Enterprises Ltd.
Image mining framework and techniques: a review
Nilanjan Dey
Department of Computer Science and Engineering,
Bengal College of Engineering and Technology,
Durgapur, West Bengal, India
Email: dey.nilanjan@ymail.com
Wahiba Ben Abdessalem Karâa
Department of CSE,
High Institute of Management,
Le Bardo, Tunis, Tunisia
Email: wahiba.abdessalem@isg.rnu.tn
Sayan Chakraborty*
Department of Computer Science and Engineering,
Bengal College of Engineering and Technology,
Durgapur, West Bengal, India
Email: sayan.cb@gmail.com
*Corresponding author
Sukanya Banerjee
Department of Computer Science and Engineering,
JIS College of Engineering,
Kalyani, West Bengal, India
Email: banerjee.sukanya7@gmail.com
Mohammed A.M. Salem
Faculty of Computer and Information Sciences,
Ain Shams University,
Abbassia, Cairo, Egypt
Email: salem@cis.asu.edu.eg
Ahmad Taher Azar
Faculty of Computers and Information,
Benha University,
Qalyubia, Egypt
Email: ahmad_t_azar@ieee.org
46 N. Dey et al.
Abstract: Image mining refers to a data mining technique where images are
used as data. It supports a large field of applications like medical diagnosis,
agriculture, industrial work, space research and obviously the educational field.
The image mining technique can extract knowledge and exciting patterns
which are not stored in the database by analysing the images using various
tools. The new era of advanced technology and high storage capability supports
the growth of large and detailed image database. This review paper presents a
detailed view on the existing research works in the area of image mining and
also summarised the different techniques used.
Keywords: image mining; object recognition; image classification; image
indexing and retrieval; association rule mining; image clustering; image mining
frameworks; neural network.
Reference to this paper should be made as follows: Dey, N., Karâa, W.B.A.,
Chakraborty, S., Bnaerjee, S., Salem, M.A.M. and Azar, A.T. (2015)
‘Image mining framework and techniques: a review’, Int. J. Image Mining,
Vol. 1, No. 1, pp.45–64.
Biographical notes: Nilanjan Dey is an Assistant Professor in the Department
of Computer Science and Engineering in Bengal College of Engineering and
Technology, Durgapur, West Bengal, India. He is a PhD Scholar of Jadavpur
University, Electronics and Telecommunication Engineering Department,
Kolkata, India and also holds an honorary position of Visiting Scientist at
Global Biomedical Technologies Inc., CA, USA. He is the Managing Editor of
International Journal of Image Mining (IJIM), Inderscience (ISSN 2055–6039)
and is the Regional Editor Asia of International Journal of Intelligent
Engineering Informatics (IJIEI), Inderscience (ISSN 1758–8723). His
research interests include: medical imaging, soft computing, data mining,
machine learning, information hiding, security, computer aided diagnosis, and
atherosclerosis. He has applied for a patent, has four books (including three
edited books), 12 book chapters and almost 100 international conferences and
journal papers.
Wahiba Ben Abdessalem Karâa completed her PhD in Computer Science,
University of Paris, VII Jussieu, France. Currently, she is an Assistant
Professor at the High Institute of Management of Tunis, Dept. of Computer
Science Applied to Management, Tunisia. She has more than 50 research
papers in various reputed journals and conferences.
Sayan Chakraborty is currently working as an Assistant Professor in
Department of Computer Science and Engineering, Bengal College of
Engineering and Technology, Durgapur, India. He completed his MTech and
BTech from Department of CSE, JIS College of Engineering, Kalyani,
West Bengal, India. He has around 28 research papers in various international
journals and conferences. Biomedical image processing, watermarking and
meta-heuristics are the key areas which he is currently working on.
Sukanya Banerjee is an MTech Scholar in Department of Computer Science
and Engineering, JIS College of Engineering and Technology, Kalyani, Nadia.
She completed her BTech from M.C.E.T., at Murshidabad, West Bengal. She
has two research papers in international journals and conference. She is
currently working on medical image processing and watermarking.
Image mining framework and techniques 47
Mohammed A.M. Salem is an Assistant Professor in the Faculty of Computer
and Information Sciences, Ain Shams University in Cairo, Egypt since 2009.
He received his PhD degree from Humboldt-University in Berlin, Germany in
November 2008, in the topic of multiresolution image segmentation. His
research interests include: robot vision, multimedia, signal processing,
multiresolution analysis, and wavelet transform. He has published a book and
more than 40 papers in the fields of image/video processing.
He has gained teaching experience and supervised different Master’s and PhD
students through his position in Ain-Shams University in Cairo and
Humboldt-University in Berlin. He used to teach the courses of mathematics
for undergraduate students and course of image processing and visual serving
for post graduate students.
Ahmad Taher Azar received his MSc in System Dynamics (2006) and PhD in
Adaptive Neuro-Fuzzy Systems (2009) from the Faculty of Engineering, Cairo
University (Egypt). He is currently an Assistant Professor in the Faculty
of Computers and Information, Benha University, Egypt. He is the
Editor-in-Chief of two journals published by IGI Global, USA such as
International Journal of System Dynamics Applications (IJSDA). He is an
Associate Editor of IEEE Trans. Neural Networks and Learning Systems. He
has worked in the areas of system dynamics, soft computing and modelling in
biomedicine and has authored/co-authored over 80 research publications.
1 Introduction
Image mining aims to extract relationships and patterns which are not explicitly stored in
database from raw data images. Image mining is a well structured technique based on
data mining, artificial intelligence, machine learning, image retrieval, image processing,
computer vision and database etc. Image mining’s capability of discovering useful image
patterns opens various research fields to new frontiers. Mining large collection of images,
and combined data mining of large collections of images with associated alphanumeric
data are the two important themes of image mining. The main reason behind the
increasing popularity of image mining is its capability to infer knowledge from the image
data automatically. Raw images or image sequences with low level pixel representation is
processed efficiently and effectively to extract the high level objects and their
relationship from those images. Image mining is still at the experimental stage. It can be
considered as an efficient hybridisation of image processing and data mining concepts to
extract the useful knowledge. In the proposed framework of Zhang et al. (2001) on data
mining, only four levels of information was extracted, they were:
a pixel level
b semantic concept level
c object level
d pattern and knowledge level.
48 N. Dey et al.
High-dimensional indexing and retrieval techniques were used to maintain the data flow
within various levels. They also increased amount of image data which made the data
mining technique more demanding with its unique features. Various application domains
(Ping and Yueshun, 2009) of image mining include natural scene recognition, remote
sensing, weather forecasting, criminal investigation, image segmentation, image
watermarking (Dey et al., 2012a, 2012b, 2011) etc.
One of the well-known techniques (Pal et al., 2013; Dey et al., 2013a, 2013b, 2013e)
that have been implemented in image processing (Dey et al., 2012c, 2012d, 2015) for
information security is watermarking (Bhattacharya et al., 2012; Dey et al., 2014, 2013c).
Watermarking refers (Dey et al., 2012e, 2012f; Chakraborty et al., 2012) to hiding or
embedding any message inside a signal (Dey et al., 2012g, 2012h), image or video. In
image watermarking (Dey et al., 2012i, 2012j; Chakraborty et al., 2013) a data or
message is taken, then embedded inside an image. This technique is known as watermark
embedding (Dey et al., 2013c). Following the retrieval of embedded or watermarked
image, the watermark is extracted from the watermarked image in order to collect the
hidden information. Watermarking can be reversible or non-reversible. Image
watermarking (Dey et al., 2013d) can also be categorised into blind and non-blind
watermarking (Dey et al., 2012k). Regarding the operations on images, image mining
differs from those traditional operations of image processing and computer vision in the
way of their working technique on images. Image mining works on large collection of
images whereas most of the previously mentioned techniques work on a single image.
The main aim of image mining is to extract some relevant information and significant
patterns from the collection of existing image database and related alphanumeric data.
The important activities in image mining (Cosa et al., 2002) are searching and retrieval of
images, based on the features and similarity of a given input query image from the image
database. There are several image mining tools available such as iARM, CAViz, web
image-gathering task, SVM classifier, B2S, DisIClass, MetaSEEk, PLSA, fully
automated age estimation engine (Devsena et al., 2011), QBIC, Photobook, SWIM,
Virage, Visualseek, Netra, MARS and so on. Now-a-days, image mining has become be
an important research topic due its applications in various areas such as medical imaging,
weather forecasting, management of earth’s resources, forest fires, criminal investigation
etc.
Image mining provides a framework that uses the raw format images stored in the
database which cannot be used directly. To use them in high-level modelling they must
be processed first. An image mining technique is considered as a good technique if it
supports fully user interaction during retrieving the patterns and knowledge from the
collection of huge image (Bach et al., 1996) database. The following functions are
performed in image mining, they are: image storage, image processing, feature
extraction, image indexing and retrieval, patterns and knowledge discovery. The two
kinds of frameworks of image mining are
1 function driven framework: which focused on different modules component and
their functionalities?
2 information driven framework: that provided a hierarchical structure of levels and
the data needed into all the levels.
Image mining framework and techniques 49
Our present work provides an overall review on such existing image mining frameworks,
techniques and further describes their attributes, features, advantages, disadvantages etc.
Section 2 provides the literature review of the previous works done in this area. Various
image mining techniques are illustrated in Section 3. Section 4 presents an overview on
the image mining frameworks. Paper concludes in Section 5.
2 Review on image mining
Fay et al. (2003) developed a system for multisensory image fusion and interactive
mining. This system was dependent on neural models of colour vision processing,
learning and pattern recognition. They also had add-on modules which performed
image conditioning, image fusion, extraction of context features and interactive image
mining. All of these modules were combined together to create a work flow which
enabled a user to create vector products of foundation features (e.g., roads, rivers, and
forests). They also highlighted the target detections from raw multisensory or
multispectral imagery. In this image mining technique, multispectral imagery modified
by simulated environmental conditions was not addressed. In order to process large
amounts of remote sensing image data, Daschiel et al. (2005) developed the prototype
model of information mining system. It consisted of both an online interface as well as an
offline part. The offline part dealt with the generation of features relating to image
mining like data reduction, compression, unsupervised content index and the absorption
of catalogue entry.
Users can collect information from a vast amount of data which is present
in the WWW on demand by using various image mining techniques. The digitalised
image which is obtained from the web is relevant in the real world. Sometimes real world
images can differ from the obtained results due to the various classification/recognition
techniques used. Morsillo et al. (2008) proposed a technique which provided more
accurate visualisation of objects by reducing the noisy search. This model combined both
generative and discriminative elements to perform an efficient retrieval of web images. It
successfully worked on semi supervised machine learning technique. Zhan et al. (2009)
devised the relation between the two main characteristics of web image i.e., visual
feature cluster and keyword, using multi-mode association rule.
Image classification is an active research area in image mining. There exists several
mining algorithm for retrieving information from the web. The quality of a good
algorithm can be determined by the process, which semantically extracts the images from
the database. Zhu et al. (2009) developed a better nonlinear algorithm which classified
the problem depending upon the distance between training and test manifold. It also
reduced the dimensionality and complexity. Later, Zhan et al. (2009) introduced a search
technique which helped us to understand sensitive Markov stationary feature (C-MSF)
after getting information from the relevant images. It represented a random walk with
restart (RWR) algorithm on images where special cooccurrence and information were
integrated. They were transformed into a classified form with the help of an SVM
classifier.
50 N. Dey et al.
Image mining helps a lot in magnetic resonance imaging (MRI) of human brain in
medical field. In the field of neurology and neurocognitive study, the clustering of
Corpus Callosum (Fatma, 2012) in midsagittal is a very critical task because of the size
and structure of Corpus Callosum. The size and structure of Corpus Callosum (Elsayed
et al., 2010; Fatma, 2012) also varies depending upon the age, sex, neurode generative
diseases and lateralised behaviour of different people. The segmentation of Corpus
Callosum during MR imaging of the brain is a very complex task in image mining.
The method proposed by Rajendran and Madheswaran (2010a) was capable of
detecting the tumour from the CT scan report of the brain by removing all other
inconsistencies from the image report. The accuracy of detecting tumour from images
using this method was much better than other techniques. Sheela and Shanthi (2007)
devised an image mining technique to identify the normal and abnormal images of brain
which led to identification of brain diseases from the MRI of abnormal tissues. Rajendran
and Madheswaran (2010b) developed a system successfully detected brain tumour from
CT scan report by pre-processing extraction of features, association rule and hybrid
classifier. In this study, by using median filtering and canny edge detection method, the
pre-processing technique extracted the edge features of the scanned images.
In association rule mining, frequent pattern tree (FP-tree) detects various patterns
generated in CT scan image report. It provides far more accurate result than any other
existing classification methods. Hybrid method is something that combines both the
mining approaches which enhances the efficiency compared to any traditional methods.
Mohan and Kannan (2010) provided a system that classified and retrieved the image by
colour. It made the process fully interactive for the user. This technique used some steps
for gathering information like colour image classification, pre-processing, pre clustering,
texture feature extraction, similarity comparison and selection of neighbouring target
image. Dubey (2010) introduced the technique based on colour histogram and image
texture. The resulting image was generated after querying necessary images.
Images can be differentiated on the basis of colour distribution by histogram method.
The images with similar colour distribution may not be semantically associated with the
images which were retrieved by global colour histogram. In this regard, Silakari et al.
(2009) developed a system which used colour moment and block truncation coding
(BTC) for retrieving the features from image database. For image database clustering
purpose K-map clustering algorithm was used. Such methods may be applied on the
different colour spaces as, RGB, HSV, and others.
3 Image mining techniques
The techniques which were used by early image miners prior to the invention of
suitable framework include pattern recognition, image indexing and retrieval, image
classification, image clustering, association rule mining, and neural network. In the
following, is a survey on these techniques? The techniques are classified on five levels of
information and the associated image or data mining operations. These levels (from top
to bottom) are:
Image mining framework and techniques 51
a knowledge extraction level
b patterns and inter-image relations level
c semantic concept level
d region, objects, or visual patterns level
e pixel level.
3.1 Object recognition
One of the key areas of image mining is object recognition, which operates data on
patterns and inter-image relations level. It finds the object relevant to the real world, from
the image by processing the provided object models. It is also known as supervised
labelling method. The system has four parts, they are:
a feature detector
b model database
c hypothesiser
d hypothesis verifier.
In 2000, Jeremy and Bonet (2000) proposed a system to find out a specific known object
in the image, which applied image processing operations on the set of ‘characteristic
maps’. In Burl et al. (1999) employed learning techniques to generate recognisers
automatically. In this work, classified examples were used to capture the domain
knowledge implicitly. Later in 2001, Gibson et al. (2001) developed an optimal
FFT-based mosaicing algorithm to find common patterns in images. The results of this
work showed that the system worked well on various kinds of images.
3.2 Image retrieval
Image retrieval refers to the process of retrieving a particular image from a large database
using data mining. Retrieval of images in image mining (Tahoun et al., 2005) is done
based on some requirement specification. There are three levels of requirement
specifications and the complexity also increase with the levels.
a level 1 retrieve the image based on some basic features of images such as texture,
colour, shape or image elements’ spatial location
b level 2 is based on image retrieval which derives the logical features such as
individual objects or persons from images
c level 3 is based on image retrieval by abstract attributes which involves a high level
reasoning in order to obtain the meaning of the objects or scenes illustrated.
Kazman and Kominek (1993) introduced three query schemas to retrieve image
information. They were
52 N. Dey et al.
a query by description
b query by associate attributes
c query by image content.
Query by associate attributes refers to the technique of taking the conventional table
structure to tailor which fulfils the purpose of image needs. Query by description means
the method that uses description along with each image, through which the user can
locate the images interested. The image description is often referred as label or keyword.
With the emergence of large-scale image repositories, the problems of vocabulary and
non-scalability caused by manual operation have become more pronounced. Hence,
content-based image retrieval (CBIR) was proposed to overcome these difficulties.
IBM’s QBIC system (Flickner et al., 1995) could retrieve image description by any
combination of colour, texture and shape as well as text keyword. This system may be
one of the popular systems amongst all other image content retrieval frameworks. It uses
R*-tree indexes to improve efficiency. Image retrieval operates data on semantic concept
level, region, objects, visual patterns level and pixel level.
3.3 Image indexing
Apart from focusing on the information requirements at various levels, it is also
important to provide support for the retrieval of image data with a fast and efficient
indexing scheme. On the contrary, the image database to be searched is too large and the
feature vectors of images are of high dimension (in the order of 102) which increases the
search complexity. To reduce such complexity reducing dimensionality or indexing high
dimensional data can be used. Image indexing handles data and images in region, objects
and visual patterns level. Reducing the dimensions can be accomplished using two
well-known methods:
a the singular value decomposition (SVD) update algorithm
b clustering.
Although, the best way to reduce complexity is to perform appropriate multi-dimensional
indexing after performing dimension reduction, which provides non-Euclidean (Rui and
Huang, 1997) similarity measures.
Lin et al. (1994) introduced an efficient technique of colour indexing for retrieving
similar type data. In this work, they increased the search time as the size of the database
increased. In 2001, Tan et al. (2001) proposed a multi-level nested R-tree index which
retrieved the structure efficiently and effectively. It helped to select appropriate technique
and also helped to design new technique by prolife the retrieval process. This process
helped to evaluate the performance of colour-spatial retrieval techniques, which led to the
selection of a suitable new technique.
3.4 Image classification and clustering
Image classification and clustering refers to the method of arranging the images into
clusters which may be done in a supervised or unsupervised way. In supervised
classification, the problem is to classify a newly encountered image from a collection
of given pre-classified images. Whereas, in unsupervised classification (or image
Image mining framework and techniques 53
clustering), without any previous knowledge the unlabeled similar type of images are
grouped together which leads to cluster generation. Clustering the images based on their
content is an important and equally challenging task to infer information from the huge
collection of images. This technique is more focused on the levels of inter-image
relations, semantic in an image, and regions. However, this technique may operate on the
large raw data.
Uehara et al. (2001) discovered the method of grouping a set of images based their
low-level visual features. This method also used a binary Bayesian classifier which
classified the vacation images into indoor and outdoor categories. The existing statistical
parameter was updated using an unsupervised technique for a maximum likelihood (ML)
classifier. As a result, a new image lacking corresponding statistical parameter demanded
the analysis of corresponding training set.
Wang and Li (1997) proposed an image-based classification method of objectionable
websites (IBCOW) which classified the websites to detect if that website was objection
enable or based on image content. The early stage of mining process is Image clustering.
Important attributes for clustering are texture, colour and shape of a particular image.
They can be used separately or in combination. Several clustering techniques
are available such as: partition-based algorithms, hierarchical clustering algorithms,
mixture-resolving and mode-seeking algorithms, nearest neighbour clustering, fuzzy
clustering, evolutionary clustering approaches etc. The abstract features by cluster can be
recognised by the domain expert following the image clustering.
3.5 Association rule mining
Ordonez and Omiecinski (1998) discussed an algorithm for image mining association
rules. This algorithm reduced I/O and CPU overhead and operated data or images on
region, objects and visual pattern level. They also built the data mining system on the top
of CBIR system. This algorithm first segmented images into blobs. Then identified and
labelled objects present in the images. Later, similarity measurement was done on those
images. The value of similarity measurement being one indicated perfect match on all
desired features, whereas zero similarity measurement value referred to the worst match
possible on those desired features. To interpret the association rules, this process also
provided the auxiliary images with identified objects.
Data mining algorithm was applied to produce object association rules. Priyatharshini
and Chitrakala (2013) described the method of using association rule in case of image
retrieval. According to this method, for each query image, all association rules which
used the query image as the antecedent (A) must be found. The consequent (B) were the
candidate images for retrieval procedure. Afterwards, those candidate images were
ranked according to their confidence value. The algorithm also mentioned the support
value of rule A B being greater than A ⇒ C if B was a subset of C. If the candidate
image set was empty or consisted of less no. of images than it should be present then the
system picked several images randomly from the database which would give every image
a chance to establish the association rules. Deshpande (2011) presented a data mining
technique for finding image content-based association rule. The purpose of this
experiment was to do feasibility study of data mining approaches based on image
content.
54 N. Dey et al.
The frequent item set discovered by traditional association rule algorithm using
iteration, needed large calculation. This issue demanded a simpler approach for image
mining (Jain et al., 2013). Thus, the technique of image mining (Banda et al., 2014; Chen
and Mei, 2014) was divided into four important phases: image pre-processing, feature
extraction, conversion of image database to transaction database, and applying
association rule mining (Wang et al., 2014; Herold et al., 2011; Khodaskar and Ladhake,
2014) to this transaction database. The proposed new association rule algorithm
(Deshpande, 2010) reduced the number of scans for a priori algorithm. This algorithm
was described in four steps. In the first step the transaction database was transformed into
Boolean matrix. In the second step, frequent 1 item set L1 was generated. The Boolean
matrix was pruned by deleting some rows and columns, in the third step. In the last step,
frequent k item sets Lk(k>1) were generated.
4 Image mining framework
There are two different frameworks of image mining (Datcu and Seidel, 2000):
1 function driven framework
2 information driven framework.
Most of the existing image mining system architectures fall under the function driven
framework. However, function driven framework is not a generalised framework. It can
be application oriented or organisation oriented. Datcu and Seidel (2000) introduced
function driven framework for intelligent satellite mining system. The function driven
framework for the multimedia miner was proposed by Zaiane and Han (1998). The
advantage of this framework was it could organise and clarify the different tasks to be
performed in image mining, but on the contrary, it was unable to differentiate levels of
vital information representation to perform meaningful image mining. This drawback
was fixed in the information driven framework.
Zhang et al. (2001) provided information driven framework for the image mining,
representing different levels of information. This framework had four pixel levels
a object level
b semantic level
c pattern level
d knowledge level.
Pixel level was the lowest level in any image mining system. It worked with the raw
information about image such as image pixels and some basic image features such as
colour, texture, and shape. It was capable of answering queries like ‘retrieve the image
with red colour’. But it could not solve queries such as ‘retrieve the image of girl’. Object
level was capable of retrieving the images for such queries. It dealt with object
information based on the primitive features in the pixel level. Object recognition assigned
correct labels to a single region or set of regions. But still it could not retrieve images for
queries such as ‘image with sad faces’. The third logical concept level generated
high-level semantic concepts from the known objects to answer such queries. These three
Image mining framework and techniques 55
levels were useful for information retrieval from the image to mine it. It supported the
entire information requirement within the image mining framework.
4.1 Function-driven frameworks
Datcu and Siedel (2000) proposed an intelligent satellite mining system that had two
modules:
a a pre-processing, data acquisition and archiving system which was required to
extract the information from the image, database of raw images, and retrieval of
image
b an image mining system to help the users to understand the detailed information of
image and detect relevant information.
Similarly, the multi media niner (Zaiane and Han, 1998) included four major
components:
a image excavator which retrieved the image and videos from the existing multimedia
database
b a pre-processor which extracted the features of the images and store the processed
data into the database
c a search kernel to generate the result depends upon the query from the image and
video database
d the discovery modules such as classifier, association and characteriser perform
image information mining routines to generate the underlying patterns and
knowledge within the images
4.2 Information-driven frameworks
The function-driven framework performed the image mining by clear decomposition and
neat arrangement of different roles and tasks. But it was unable to describe the
information representation at different levels which was needed before any mining task.
Zhang et al. (2001) proposed an information-driven system which dealt with this
problem. The key attributes of this system were:
a pixel level which was the lowest level of image, consisted of the raw information
about image such as image pixels and the image features such as texture, colour and
shape
b object level referred to the object or region of information based on the result of
primitive features in the pixel level
c semantic concept level created high-level semantic concepts from the known objects
of the knowledge domain.
d pattern and knowledge level extracted the patterns and knowledge from the domain
related alphanumeric data and the logical concepts obtain from image data.
56 N. Dey et al.
Table 1 Image mining review in tabular form
Sl. no. Year of pub. Paper title Authors Short description
1 1993 Information organization in
multimedia resources
Kazman and Kominek
(1993)
Introduced three query schemas to retrieve image information. They were
a query by description
b query by associate attributes
c query by image content.
Query by associate attributes referred to the technique of taking the conventional table
structure to tailor which fulfils the purpose of image needs
2 1994
The TV-tree: an index structure
for high-dimensional data
Lin et al. (1994) Introduced an efficient technique of color indexing to retrieve similar type data.
The search time increased as the size of the database increased.
3 1995 Query by image and video
content: the QBIC system
Flickner et al. (1995) Proposed IBM’s QBIC system that could retrieve image description by any
combination of color, texture and shape as well as text keyword. This system may be
one of the popular systems amongst all other image content retrieval frameworks. It
used R*-tree indexes to improve efficiency.
4 1995 A scheme for visual features
based image retrieval
Zhang and Zhong
(1995)
Proposed the use of self-organisation map (SOM) neural nets which was the tool for
constructing the tree indexing structure. Advantages of using SOM were its
unsupervised learning ability and dynamic clustering nature.
5 1996
Image search engine: an open
framework for image management
Bach et al. (1996) Proposed a system that performed the following functions in image mining such as:
image storage, image processing, feature extraction, image indexing and retrieval,
patterns and knowledge discovery. The two frameworks of image mining were
1 function driven framework: which focused on different modules component as well
the functionalities
2 information driven framework: provided a hierarchical structure of levels and the
data needed into all the levels.
6 1996
Automatic detection of diabetic
retinopathy using an artificial neural
network: a screening tool
Gardner and Keating
(1996)
Applied artificial neural network (ANN) on image mining which provided an
automated approach of fund us image analysis by computer. This process improved the
efficiency of the assessment work of the image by offering an immediate classification
of the fund us of the patient at the time of acquisition of the image
7 1997 System for screening objectionable
images using daubechies’ wavelets and
color histograms
Wang and Li (1997) Proposed an image-based classification method of objectionable websites (IBCOW)
which classified the websites to detect if that website was objection enable or based on
image content
8 1997 Image retrieval: past,
present and future
Rui and Huang (1997) Performed appropriate multi-dimensional indexing after dimension reduction which
provided non-Euclidean similarity measures
Image mining framework and techniques 57
Table 1 Image mining review in tabular form (continued)
Sl. no. Year of pub. Paper title Authors Short description
9 1998 Image mining: a new
approach for data mining
Ordonez and
Omiecinski (1998)
Proposed an algorithm for mining that reduced I/O and CPU overhead. This led to
generating new data mining sy stem on the top of content-based image retrieval (CBIR)
system.
10 1998 Mining multimedia data Zaiane and Han
(1998)
Function driven framework for the multimedia miner was proposed. The features of
this framework was that it could organise and clarify the different tasks to be performed
in image mining, but it remained unable to differentiate levels of necessary image
information representation to perform meaningful mining.
11 1999 Mining for image content Burl et al. (1999) Proposed learning techniques to generate recognisers automatically. Classified
examples were used to capture the domain knowledge implicitly
12 2000 Image preprocessing for rapid
selection in pay attention mode
Jeremy and Bonet
(2000)
Proposed a system to find out a specific known object in the image.
13 2000
Image information mining: exploration
of image content in large archives
Datcu and Seidel
(2000)
Proposed an intelligent satellite mining system that included pre-processing, data
acquisition and archiving system which was required to extract the information from
the image, database of raw images, and retrieval of image.
14 2001
Retrieving similar shapes effectively
and efficiently, multimedia tools and
applications
Tan et al. (2001) Propose a multi-level nested R-tree index which retrieves the structure efficiently and
effectively. It helped to select appropriate technique and also helped to design new
technique by proliferate of the retrieval process.
15 2001 A computer-aided visual
exploration system for knowledge
discovery from images
Uehara et al. (2001) Discovered the method of grouping a set of images based on low-level visual features.
This method also used a binary Bayesian classifier which classified the vacation images
into indoor and outdoor categories.
16 2001
Intelligent mining in image databases,
with applications to satellite imaging
and web search, data mining and
computational intelligence
Gibson et al. (2001) Developed an optimal FFT-based mosaicing algorithm to find common patterns in
images and showed that it works well on various kinds of images.
17 2001
An information-driven framework for
image mining
Zhang et al. (2001) New information driven framework was proposed which fixed the problems generated
during function driven framework.
18 2002 Image mining by content Conc et al. (2002) Proposed a technique which helped to extract vital information from any image.
19 2003 Image fusion & mining tools for a
COTS environment
Fay et al. (2003) Proposed a system for multisensory image fusion and interactive mining. This system
was dependant on neural models of colour vision processing, learning and pattern
recognition.
20 2005
ARIRS: association rule based image
retrieval system
Yi et al. (2005) Introduced the technique to use association rule for image retrieval.
58 N. Dey et al.
Table 1 Image mining review in tabular form (continued)
Sl. no. Year of pub. Paper title Authors Short description
21 2005 Information mining in remote sensing
image archives: system evaluation
Daschiel and Datcu
(2005)
In order to process large amounts of remote sensing image data, they developed the
prototype model of information mining system. It consisted of both an online interface
as well as an offline part.
22 2007 Image mining techniques for
classification and segmentation of
brain MRI data
Sheela and Shanthi
(S2007)
Devised an image mining technique to identify the normal and abnormal images of
brain in order to identify any brain diseases from the MRI of abnormal tissues
23 2008 Mining the web for visual concepts Morsillo et al. (2008) Proposed a technique which provided more accurate visualisation of objects by
reducing the noisy search. This model combined both generative and discriminative
elements to perform an efficient retrieval of web images.
24 2009
Multi-modal mining in web image
retrieval computational intelligence
and industrial applications
Zhan (2009) Devised the relation between the two main characteristics of web image i.e. visual
feature cluster and keyword, using multi-mode association rule.
25 2009 Image classification
approach based on manifold
learning in web image mining
Zhu et al. (2009) Developed a better nonlinear algorithm which classified the problem depending upon
the distance between training and test manifold. It also reduced the dimensionality and
complexity
26 2009
Web image mining using concept
sensitive Markov stationary features
Zhang et al. 2009) Introduced a search technique which helped us to understand sensitive Markov
stationary feature (C-MSF) after getting information from the relevant images. It
represented a random walk with restart (RWR) algorithm on images where special
co-occurrence and information were integrated. They were transformed into a classified
form with the help of SVM classifier.
27 2009 Color image clustering using
block truncation algorithm
Silakari et al. (2009) Developed a system which used colour moment and block truncation coding (BTC) for
retrieving the features from image database. For image database clustering purpose
K-map clustering algorithm was used.
28 2010
Novel fuzzy association rule image
mining algorithm for medical decision
support system
Rajendran and
Madheswaran (2010b)
Introduced a search technique which helped to understand sensitive Markov stationary
feature (C-MSF) after getting information from the relevant images. It represented a
random walk with restart (RWR) algorithm on images where special co-occurrence and
information are integrated. They are transformed into a classified form with the help of
an SVM classifier.
29 2010
Hybrid medical image classification
using association rule mining with
decision tree algorithm
Rajendran and
Madheswaran (2010a)
Proposed a method where the tumour could be detected from the CT scan report of the
brain by removing all other inconsistencies from the image report. The accuracy of
detecting tumour from images using this method was much better than other
techniques.
Image mining framework and techniques 59
Table 1 Image mining review in tabular form (continued)
Sl. no. Year of pub. Paper title Authors Short description
30 2010 Color image classification
and retrieval using image
mining techniques
Mohan and Kannan
(2010)
A new technique of colour image classification and retrieval was proposed to improve
user interaction with image retrieval systems by fully exploiting the similarity
information.
31 2010
Image mining using content based
image retrieval system
Dubey (2010) Devised the technique based on colour histogram and image texture. The resulting
image was generated after querying necessary images.
32 2011
Association rule mining based on
image content
Deshpande (2011) Presented a data mining technique for finding image content based association rule. A
feasibility study of data mining was done based on image content.
33 2011
An experiential survey on image
mining tools, techniques and
applications
Devasena et al. (2011) Provided fully automated age estimation engine QBIC, Photobook, Swim, Virage,
Visualseek, Netra, MARS etc.
34 2005 Robust content-based image
retrieval system using multiple features
representations
Tahoun et al. (2005) Comparison between the combination of wavelet-based representations of the texture
feature and the colour feature with and without using the colour layout feature was
done.
35 2013 Image mining for image retrieval using
hierarchical k-means algorithm
Jain et al. (2013) Image retrieval was done using K-means algorithm.
36 2014 Big data new frontiers:
mining, search and management of
massive repositories of solar image
data and solar events
Banda et al. (2014) Used solar image data and events, and the process to use big data methodologies
37 2014 Mining weakly labeled web
facial images for search-based
face annotation
Wang et al. (2014) Proposed a method of image mining using weak facial images for face annotation.
38 2014 Mining mid-level features for
image classification
Fernando et al. (2014) Proposed an image classification technique using mid-level features.
39 2014
Toward dynamic scene understanding
by hierarchical motion pattern mining
Song et al. (2014) Used hierarchical motion pattern mining for understanding dynamic scene.
40 2011 Multivariate image mining Herold et al. (2011) Discussed about the multivariate image mining.
41 2014 Image mining: an overview
of current research
Khodaskar and
Ladhake (2014)
Illustrated the current research status on the domain of image mining.
42 2014
Mining frequent items in data stream
using time fading model
Chen and Mei (2014) Introduced new image mining frequent items, during data stream.
60 N. Dey et al.
The four levels were generalised further into two layers: lower layer comprised with the
pixel level and the object level, while the upper layer was concerned with pattern and
knowledge level as well as the semantic concept level. The lower layer consisted of raw
and extracted image information and performed the image processing, images analysis
and recognition. Operations such as semantic concept generation, knowledge discovery
from image database were caused by the higher layer. The main differences between two
layers are the upper layer information was more logical and meaningful than that of
lower level information.
5 Comparative study
This paper discussed and compared different image mining techniques and also discussed
about various image mining frameworks. The discussion and overview of all such
techniques and frameworks, helped to establish a comparative study among the existing
image mining methods. This paper also provided a comparative approach between the
image mining (Fernando et al., 2014) methods in tabular form (see Table 1). Although,
no method or framework was established as a superior than others. This paper was more
focused towards different attributes of various methods than claiming one as the superior
to others
6 Conclusions
The image mining task on image datasets majorly deals with classification, clustering,
and/or mining of knowledge from images using association rules and neural network. It
can be used to group the images on remote sensing, world wide web, medical diagnosis,
efficient retrieval of images, or to extract hidden meaningful information from image
datasets which is not explicitly available from image sources. Hence, this review paper
will help us in selecting an appropriate image mining technique among all the available
techniques. This paper still remains pilot in nature and requires further validation. Future
work may include discussion about new image mining methods and the updated
frameworks, also comparing them with previously discussed methods.
References
Bach, J.R., Fuller, C., Gupta, A., Hampapur, A., Horowitz, B., Humphrey, R., Jain, R. and Virage
CFS (1996) ‘Image search engine: an open framework for image management’, Storage and
Retrieval for Image and Video Databases, SPIE, Vol. 6, No. 76, pp.76–87.
Banda, J.M., Schuh, M.A., Angryk, R.A., Pillai, G.K. and McInerney, P. (2014) ‘Big data new
frontiers: mining, search and management of massive repositories of solar image data and
solar events’, Advances in Intelligent Systems and Computing, Vol. 241, No. 2, pp.151–158.
Bhattacharya, T., Dey, N., Chaudhuri, S.R.B. (2012) ‘A session based multiple image hiding
technique using DWT and DCT’, International Journal of Computer Applications, Vol. 38,
No. 5, pp.18–21.
Burl, M.C., Fowlkes, C. and Rowden, J. (1999) ‘Mining for image content’, Systemics,
Cybernetics, and Informatics and Information Systems: Analysis and Synthesis, pp.126–134,
Orlando, FL.
Image mining framework and techniques 61
Chakraborty, S., Maji, P., Pal, A.K., Biswas, D. and Dey, N. (2012) ‘Reversible color image
watermarking using trigonometric functions’, International Conference on Electronic
Systems, Signal Processing and Computing Technologies, Nagpur, pp.105–110.
Chakraborty, S., Samanta, S., Mukherjee, A., Dey, N. and Chaudhuri, S.S. (2013) ‘Particle swarm
optimization based parameter optimization technique in medical information hiding’, IEEE
International Conference on Computational Intelligence and Computing Research (ICCIC),
pp.1–6.
Chen, L. and Mei, Q. (2014) ‘Mining frequent items in data stream using time fading model’,
Information Sciences, Vol. 257, pp.54–69.
Conc, A., Mathias, E. and Castro, M.M. (2002) ‘Image mining by content’, Expert Systems with
Applications, Vol. 23, No. 4, pp.377–383.
Das, P., Munshi, R. and Dey, N. (2012) ‘Alattar’s method based reversible watermarking technique
of EPR within heart sound in wireless telemonitoring’, Intellectual Property Rights And
Patent Laws, IPRPL-2012, Jadavpur University, August 25, 2012.
Daschiel, H. and Datcu, M. (2005) ‘Information mining in remote sensing image archives: system
evaluation’, IEEE Transactions on Geoscience and Remote Sensing, Vol. 43, No. 1,
pp.188–199.
Datcu, M. and Seidel, K. (2000) ‘Image information mining: exploration of image content in large
archives’, IEEE Conference on Aerospace, Vol. 3, pp.253–264.
Deshpande, D. (2011) ‘Association rule mining based on image content’, International Journal of
Information Technology and Knowledge Management, Vol. 4, No. 1, pp.143–146.
Devasena, L., Sumathi, T. and Hemalatha, M. (2011) ‘An experiential survey on image mining
tools, techniques and applications’, International Journal on Computer Science and
Engineering (IJCSE), Vol. 3, No. 3, pp.1155–1160.
Dey, N., Biswas, S., Roy, A.B., Das, A. and Chaudhuri, S.S. (2012a) ‘Analysis of
photoplethysmographic signals modified by reversible watermarking technique using
prediction-error in wireless telecardiology’, International Conference of Intelligent
Infrastructure, 47th Annual National Convention of CSI, McGraw-Hill Proceeding.
Dey, N., Das, P., Das, A. and Chaudhuri, S.S. (2012b) ‘DWT-DCT-SVD based intravascular
ultrasound video watermarking’, Second World Congress on Information and Communication
Technologies (WICT 2012), pp.224–229.
Dey, N., Das, P., Das, A. and Chaudhuri, S.S. (2012c) ‘DWT-DCT-SVD based blind watermarking
technique of gray scale image in electrooculogram signal’, International Conference on
Intelligent Systems Design and Applications (ISDA-2012), pp.680–685.
Dey, N., Das, P., Das, A. and Chaudhuri, S.S. (2012d) ‘DWT-DCT-SVD based intravascular
ultrasound video watermarking’, Second World Congress on Information and Communication
Technologies (WICT 2012), pp.224–229.
Dey, N., Das, P., Das, A. and Chaudhuri, S.S. (2012e) ‘DWT-DCT-SVD based blind watermarking
technique of gray scale image in electrooculogram signal’, International Conference on
Intelligent Systems Design and Applications (ISDA-2012), pp.680–685.
Dey, N., Das, P., Das, A. and Chaudhuri, S.S. (2012f) ‘Feature analysis for the blind-watermarked
electroencephalogram signal in wireless telemonitoring using Alattar’s method’, 5th
International Conference on Security of Information and Networks (SIN 2012), in technical
cooperation with ACM Special Interest Group on Security, Audit and Control (SIGSAC),
Jaipur, India, pp.87–94.
Dey, N., Mishra, G., Nandi, B., Pal, M., Das, A. and Chaudhuri, S.S. (2012g) ‘Wavelet based
watermarked normal and abnormal heart sound identification using spectrogram analysis’,
2012 IEEE International Conference on Computational Intelligence and Computing Research
(ICCIC), pp.1–7.
62 N. Dey et al.
Dey, N., Mukhopadhyay, S., Das, A. and Chaudhuri, S.S. (2012h) ‘Using DWT analysis of P, QRS
and T components and cardiac output modified by blind watermarking technique within the
electrocardiogram signal for authentication in the wireless telecardiology’, I.J. Image,
Graphics and Signal Processing (IJIGSP), Vol. 4, No. 7, pp.33–46, ISSN:2074–9074.
Dey, N.S., Biswas, Das, P., Das, A. and Chaudhuri, S.S. (2012i) ‘Feature analysis for the reversible
watermarked electrooculography signal using low distortion prediction-error expansion’, 2012
International Conference on Communications, Devices and Intelligent Systems (CODIS),
pp.624–627.
Dey, N.S., Biswas, Das, P., Das, A. and Chaudhuri, S.S. (2012j) ‘Lifting wavelet transformation
based blind watermarking technique of photoplethysmographic signals in wireless
telecardiology’, Second World Congress on Information and Communication Technologies
(WICT 2012), pp.230–235.
Dey, N., Chakraborty, S. and Samanta, S. (2013a) Optimization of Watermarking in Biomedical
Signal, Lambert Publication, Heinrich-Böcking-Straße 6, 66121 Saarbrücken, Germany
[ISBN-13: 978-3-659-46460-7].
Dey, N., Das, P., Biswas, D., Maji, P., Das, A. and Chaudhuri, S.S. (2013b) ‘Visible watermarking
within the region of non-interest of medical images based on fuzzy c-means and Harris corner
detection’, The Fourth International Workshop Communications Security & Information
Assurance (CSIA-2013), Springer, pp.161–168.
Dey, N., Maji, P., Das, P., Das, A. and Chaudhuri, S.S. (2013c) ‘An edge based watermarking
technique of medical images without devalorizing diagnostic parameters’, International
Conference on Advances in technology and Engineering, pp.1–5.
Dey, N., Nandi, B., Das, P., Das, A. and Chaudhuri, S.S. (2013d) ‘Retention of electrocardiogram
features insignificantly devalorized as an effect of watermarking for a multi-modal biometric
authentication system’, Advances in Biometrics for Secure Human Authentication and
Recognition, ISBN-9781466582422, pp.450.
Dey, N., Samanta, S., Yang, X-S., Chaudhuri, S.S. and Das, A. (2013e) ‘Optimization of scaling
factors in electrocardiogram signal watermarking using Cuckoo Search’, in International
Journal of Bio-Inspired Computation (IJBIC), Vol. 5, No. 5, pp.315–326.
Dey, N., Dey, G., Chakraborty, S., Chaudhuri, S.S. (2014) ‘Feature analysis of blind watermarked
electromyogram signal in wireless telemonitoring’, Concepts and Trends in Healthcare
Information System of the Annals of Information Systems Series, Vol. 16, No. 2014,
pp.205–229.
Dey, N., Dey, M., Biswas, D., Das, P., Das, A. and Chaudhuri, S.S. (2015) ‘Tamper detection of
electrocardiographic signal using watermarked bio-hash code in wireless cardiology’, Special
Issue of the International Journal of Signal and Imaging Systems Engineering, Vol. 8, No. 1,
pp.46–58.
Dey, N., Roy, A.B. and Dey, S. (2011) ‘A novel approach of color image hiding using RGB color
planes and DWT’, International Journal of Computer Applications, Vol. 36, No. 5, pp.19–24.
Dubey, R.S. (2010) ‘Image mining using content based image retrieval system’, (IJCSE)
International Journal on Computer Science and Engineering, Vol. 2, No. 7, pp.2353–2356.
Elsayed, A, Coenen, F., Jiang, C., García-Fiñana, M. and Sluming, V. (2010) ‘Corpus callosum
MR image classification’, Journal Knowledge-Based Systems, Elsevier Science Publishers B,
V. Amsterdam, The Netherlands, The Netherlands, Vol. 23, No. 4, pp.330–336.
Fatma, S.N. (2012) ‘Image mining method and frameworks’, International Journal of
Computational Engineering Research (ijceronline.com) Vol. 2, No. 8, pp.131–145.
Fay, A., Ivey, R.T., Bomberger, N. and Waxm, A.M. (2003) ‘Image fusion & mining tools for a
COTS environment’, Proceedings of the Sixth International Conference of Information
Fusion, Vol. 1, pp.606–613.
Fernando, B., Fromont, E. and Tuytelaars, T. (2014) ‘Mining mid-level features for image
classification’, International Journal of Computer Vision, Vol. 108, No. 3, pp.1–30.
Image mining framework and techniques 63
Flickner, M., Sawhney, H.S., Ashley, J., Huang, Q., Dom, B., Gorkani, M., Hafner, J., Lee, D.,
Petkovic, D., Steele, D. and Yanker, P. (1995) ‘Query by image and video content: the QBIC
system’, IEEE Computer, Vol. 28, No. 9, pp.23–32.
Gardner, G. and Keating, D. (1996) ‘Automatic detection of diabetic retinopathy using an artificial
neural network: a screening tool’. British Journal of Ophthalmology, Vol. 80, No.11,
pp.940–944.
Gibson, S., Kreinovich, V., Longpre, L., Penn, B. and Starks, S.A. (2001) Intelligent Mining in
Image Databases, with Applications to Satellite Imaging and Web Search, Data Mining and
Computational Intelligence, pp.1–20, Springer-Verlag, Berlin.
Herold, J., Loyek, C. and Nattkemper, T.W. (2011) ‘Multivariate image mining’, Wires Data
Mining Knowledge Discovery, Vol. 1, No. 1, pp.2–13.
Jain, P.M., Gawande, A.D. and Gautam, L.K. (2013) ‘Image mining for image retrieval using
hierarchical K-means algorithm’, International Journal of Research in Computer Engineering
and Electronics, Vol. 2, No. 6, pp.1–6.
Jeremy, S. and Bonet, D. (2000) ‘Image preprocessing for rapid selection in pay attention mode’,
MIT Press.
Kazman, R. and Kominek, J (1993) ‘Information organization in multimedia resources’, 11th
Annual International Conference on Systems Documentation, pp.149–162.
Khodaskar, A.A. and Ladhake, S.A. (2014) ‘Image mining: an overview of current research’,
Fourth International Conference on Communication Systems and Network Technologies
(CSNT), pp.433–438.
Lin, K., Jagadish, H.V. and Faloutsos, C. (1994) ‘The TV-tree: an index structure for
high-dimensional data’, The VLDB Journal, Vol. 3, No. 4, pp.517–542.
Mohan, V. and Kannan, A. (2010) ‘Color image classification and retrieval using image mining
techniques’, International Journal of Engineering Science and Technology, Vol. 2, No. 5,
pp.1014–1020.
Morsillo, N., Pal, C. and Nelson, R. (2008) ‘Mining the web for visual concepts’, Proceedings of
the 9th International Workshop on Multimedia Data Mining: Held in Conjunction with the
ACM SIGKDD 2008, pp.18–25.
Ordonez, C. and Omiecinski, E. (1998) ‘Image mining: a new approach for data mining’, Research
Paper, pp.1–20.
Pal, A.K., Das, P. and Dey, N. (2013) Odd-Even Embedding Scheme Based Modified Reversible
Watermarking Technique Using Blueprint, arXiv preprintarXiv:1303.5972.
Ping, D. and Yueshun, H. (2009) ‘Research of remote sensing image data mining technique based
on web’, Asia-Pacific Conference on Information Processing, Shenzhen, APCIP, Vol. 1,
pp.298–300.
Priyatharshini, R. and Chitrakala, S. (2013) ‘Association rule based image retrieval system’,
Communications in Computer and Information Science, Vol. 296, pp.17–26.
Rajendran, P. and Madheswaran, M. (2010a) ‘Hybrid medical image classification using
association rule mining with decision tree algorithm’, Journal of Computing, Vol. 2, No. 1,
pp.127–136.
Rajendran, P. and Madheswaran, M. (2010b) ‘Novel fuzzy association rule image mining
algorithm for medical decision support system’, International Journal of Computer
Applications, Vol. 1, No. 20, pp.0975–8887.
Rui, Y. and Huang, T.S. (1997) ‘Image retrieval: past, present and future’, Journal of Visual
Communication and Image Representation, Vol. 10, pp.39–62.
Sheela, L.J. and Shanthi, V. (2007) ‘Image mining techniques for classification and segmentation
of brain MRI data’, Journal of Theoretical & Applied Information Technology, Vol. 3, No. 4,
pp.115–121.
64 N. Dey et al.
Silakari, S., Motwani, M. and Maheshwari, M. (2009) ‘Color image clustering using block
truncation algorithm’, IJCSI International Journal of Computer Science Issues, Vol. 4, No. 2,
pp.31–35.
Song, L., Jiang, F., Shi, Z., Molina, R. and Katsaggelos, A.K. (2014) ‘Toward dynamic scene
understanding by hierarchical motion pattern mining’, IEEE Transactions on Intelligent
Transportation Systems, Vol. 15, No. 3, pp.1273–1285.
Tahoun, M.A., Nagaty, K.A., El-Arief, T.I. and Megeed, M.A. (2005) ‘Robust content-based image
retrieval system using multiple features representations’, Proc. of Networking, Sensing and
Control, pp.116–122.
Tan, K.L., Ooi, B.C. and Thiang, L.F. (2001) Retrieving Similar Shapes Effectively and Efficiently,
Multimedia Tools and Applications, Kluwer Academic Publishers, The Netherlands.
Uehara, Y., Endo, S., Shiitani, S., Masumoto, D. and Nagata, S. (2001) ‘A computer-aided visual
exploration system for knowledge discovery from images’, Second International Workshop on
Multimedia Data Mining (MDM/KDD’2001), pp.102–109.
Wang, D., Hoi, S.C.H., He, Y. and Zhu, J. (2014) ‘Mining weakly labeled web facial images for
search-based face annotation’, IEEE Transactions on Knowledge and Data Engineering,
Vol. 26, No. 1, pp.166–179.
Wang, J.Z. and Li, J. (1997) ‘System for screening objectionable images using daubechies’,
wavelets and color histograms’, Proceedings of the Fourth European Workshop (IDMS’97),
pp.1–12.
Yi, H., Rajan, D. and Chia, L-T. (2005) ‘ARIRS: association rule based image retrieval system’,
International Workshop for Advanced Imaging Technology (IWAIT‘05), pp.1–6.
Zaiane, O.R. and Han, J.W. (1998) ‘Mining multimedia data’, CASCON’98, pp.83–96.
Zhan, R.H.W. (2009) ‘Multi-modal mining in web image retrieval computational intelligence and
industrial applications’, PACIIA 2009, Vol. 2, pp.425–428.
Zhang, H. and Zhong, D. (1995) ‘A scheme for visual features based image retrieval’, Storage and
Retrieval for Image and Video Databases, SPIE, pp.1–7.
Zhang, J., Hsu, W. and Lee, M.L. (2001) ‘An information-driven framework for image mining’,
Database and Expert Systems Applications in Computer Science, Vol. 2113, pp.232–242.
Zhang, J., Liu, H. and Ma, S. (2009) ‘Web image mining using concept sensitive Markov
stationary features’, IEEE International Conference on Multimedia and Expo, pp.462–465.
Zhu, R., Yao, M. and Liu, Y. (2009) ‘Image classification approach based on manifold learning in
web image mining’, Advanced Data Mining and Applications, Vol. 5678, pp.780–787.
