https://doi.or g/10.31449/inf.v47i7.4583 Informatica 47 (2023) 81–90 81
A New Multimedia W eb-Data Mining Appr oach based on Equivalence Class
Evaluation Pipelined to Featur e Maps onto Planar Pr ojection
M. Ravi
1, 3
, M. Ekambaram Naidu
2
and G. Narsimha
3
1
CMR Institute of T echnology , Hyderabad-501401, T elangana, India
2
SRK Institute of T echnology , V ijayawada-521 108, India
3
Jawaharlal Nehru T ech University , Hyderabad-500085, India
E-mail: ravimogili@gmail.com, menaidu2005@yahoo.co.in, narsimha06@gmail.com
Keywords: supervised learning, information retrieval, multimedia web-databases, statistical metrics, feature extraction
Received: Dec 25, 2022
Multimedia information ar e semi-or ganized or unstructur ed information elements whose essential sub-
stance is separately or by and lar ge utilized for corr espondence. Sight and sound information mining
r ecognizes, arranges, and r ecovers important highlights fr om an assortment of media to r ecognize en-
lightening examples furthermor e, connections for information acquisition. Computer V ision (CV)-based
systems have been incr easingly popular in r ecent years, owing to the gr owing number and complexity of
datasets. In CV , finding meaningful photos in a huge dataset is a difficult task to solve. T raditional sear ch
engines r etrieve photos based on text such as captions and metadata, but this strategy can r esult in a lot of
irr elevant output, not to speak the time, effort, and money r equir ed to tag this textual data.
In this paper , we pr oposed a pipelined deep learning oriented methodology framework f or multimedia web-
data mining based on content extracted featur e maps in planner pr ojection as input. Color , textur e, form,
and other high-level pr operties of images ar e r epr esented as numerical featur e vectors. This technique is
based on the following computer vision tasks in general i.e., Image segmentation, Image classification, Ob-
ject detection etc. In or der to pr ove the computational efficiency and to validate its statistical behaviour ,
we have also pr esented the experimental evaluation on an standar d multimedia dataset. The obtained
performance r esults ar e then compar ed with some significant existing appr oaches in the terms of various
statistical measur es/parameters.
Povzetek: Pr edstavljena je metoda rudarja multimedijev z globokim učenjem, ki temelji na lastnostih vse-
bine slik. Uporablja se za različne naloge računalniškega vida, kot so segmentacija, klasifikacija in zaz-
navanje objektov . Pr eizkušena je bila na standar dnem multimedialnem naboru podatkov .
1 Intr oduction
Multimedia information mining involves identifying, cat-
egorizing, and retrieving relevant features from a variety
of media to identify educational patterns and relationships
for data acquisition. The performance of multimedia in-
formation mining is directly impacted by the level of in-
formation contained in the data, which can be recognized
during the data pre-processing stage. Data pre-processing
aims to reduce data dimensionality by preserving useful
features, which are factors (or attributes) used as input for
selection algorithms. This greatly improves runtime (train-
ing time), predictive accuracy , and result readability when
working with human interpretation and knowledge. There
are two key steps in data pre-processing: feature extrac-
tion (FE) and feature selection (FS). FS techniques are a
subset of the broader field of FE. While FE extracts mul-
tiple features from the original data to generate a dataset,
FS selects a subset of original variables to build the model.
However , identifying related features provides insights into
underlying phenomena. On the other hand, eliminating ir -
relevant variables enhances the accuracy of data classifica-
tion. Multimedia web-information mining includes various
applications such as face recognition, pattern recognition,
text detection and recognition in images and video frames,
biomedicine, emotion recognition, image retrieval and clas-
sification, and image annotation, among others.
1.1 Multimedia
Multimedia information are semi-or ganized or unstructured
information elements whose essential substance is exclu-
sively or all in all utilized for correspondence. The term
sight and sound covers a variety of media article like a mix
of text, picture, video, sound, numeric, sound, liveliness,
graphical, and straight out information on a PC show ter -
minal. In the approach of programming innovation, sight
and sound is characterized as ”a PC program comprised of
messages, realistic, sound and pictures and animations”.
T ext is made out of an enormous number of disor ganised
and nebulous characters of normal language text [22, 23].
W ritings can be inserted in recordings in two structures,

82 Informatica 47 (2023) 81–90 M. Ravi et al.
specifically , inscription and scene messages [24]. Subti-
tle text alludes to the writings overlapped during video al-
tering, while scene oriented text is the writings that nor -
mally exist during the video catches. T o manage the im-
planted content for a proper trade of data (e.g., title as
well as date of an occasion, name of the speakers), sub-
title text can be extricated straightforwardly from record-
ings and pictures since it is just positioned in comparison to
these visual representations. Sound is a sign coming about
because of motions or pressing factor varieties created by
a moving or vibrating body tempestuous liquid stream in
a flexible medium like air , water , and solids [25]. Uzun
and Sencar [26] characterized sound into discourse, mu-
sic, non-discourse as well as non-musical signals, as well
as complicated sound combinations that emer ged from a
few particular acoustic sources. Sound substance can be
naturally examined and looked inside itself. For instance,
discourse w ith in sound discovery is generally utilized for
distinguishing proof or acknowledgment, for example, sup-
porting proof assortment and diagnosing Alzheimer ’ s dis-
ease.
1.2 Challenges in multimedia data mining
There are plentiful measure of dif ficulties in the multime-
dia information mining. Information with high and com-
plex measurements experience the ill ef fects of the scour ge
of dimensionality [27]. In these wonders, the measure of
information required to help the outcomes develop dramat-
ically with dimensionality , accordingly sabotaging the pre-
sentation of information mining [28-29]. At the end of the
day , the volume of data space increments and involves the
sparsity of dependable information through an additional
measurement in the investigation information. High di-
mensional information entangle sight and sound examina-
tion, stockpiling, and recovery . The significant component
space ef fectively prompts overfitting; when the information
digging mechanism looks for the best boundaries utilizing
restricted information, it evokes both the overall examples
in the information and the commotion explicit to the infor -
mation. As far as precision, ef fectiveness, and dependabil-
ity , recovering an ideal mixed media content (e.g., image)
in a continually developing information base is confounded
in light of the trouble in acquiring a legitimate image high-
light set.
However , it has been on the whole concurred that the arti-
cle image explanation is fundamentally an order issue, most
web pictures are multi-marked. Along these lines, an im-
age can likewise mirror a few semantic ideas. Multi-name
characterization likewise has an alternate set of assessment
measures contrasted with single-name grouping. The tradi-
tional calculations for the most part change multi-name ar -
rangement issues several parallel order issues for each idea.
T wo sorts of commotion existed in multimedia information.
The main kind is identified with the foundation commo-
tion in mixed media information. Foundation commotion
can happen in recordings or sound information in light of a
few reasons, counting foundation voices recorded through
an amplifier and defective melody replicating framework.
order precision may be made even better with the deter -
mination of appropriate discriminative element gatherings.
Subsequently , a decent model with a decent FS capacity
is needed to improve both exactness and productivity , that
is a high AUC esteem with a similar chose highlight mea-
surement shows a agreeable execution with a low involved
component unit number . Assessment of highlights individ-
ually implies that the significance of each features are ex-
clusively chosen for the unmistakable includes as opposed
to considering the relationship among attributes.
1.3 Generalized methods for multimedia
web-data mining
The general methods of multimedia web-data mining can be
categorized in the following two ways i.e. (i) based on input
text word(s) query as input (ii) based on extracted feature
maps in planner projection as input.
1.3.1 Multimedia web-data mining based on input
text word(s) query as input
W ords or sentences that convey content are known as
catchphrases. They may be used to represent photos, text
archives, database entries, and W eb pages as metadata. Set-
ting up catchers for a photo allows you to retrieve, record,
sort, and see lar ge amounts of image data. Catchphrases are
utilized on the W eb in two unique manners: I) Keywords as
a scan terms for web crawlers ii) Keywords that recognize
the substance of the site. A comment is metadata appended
to text, picture, or other information. It alludes to a partic-
ular piece of the unique information or image.
Limitations in this type of methods:- The limitations in
this type of methods are as follows:-
– The assignment of envisioning picture content is pro-
foundly emotional.
– Accompany the significant list items; it very well may
be a huge number of unessential query items It might
mean that the evidenced pursuit’ s precision is low .
– The text based portrayals given by an annotator ought
to be unique in relation to the next client. For dif ferent
people, an image represents dif ferent things. It might
also imply dif ferent things to dif ferent people at dif-
ferent times.
1.3.2 Multimedia web-data mining based on
extracted featur e maps in planner pr ojection as
input
T o overcome the limitations of multimedia W eb-Data Min-
ing based on input text word(s) query as input, we can use
the extracted feature maps in planner projection as input

A New Multimedia W eb-Data Mining Approach… Informatica 47 (2023) 81–90 83
based mining. It is the use of computer vision for recov-
ering the image object. Data recovery implies the way to-
ward changing over a solicitation for data into a significant
arrangement of reference. It is an innovation that in rule
puts together computerized image files as per their visual
substance. This framework recognizes the extraordinary
locales present in a image dependent on their similitude in
shading, surface, shape, and so on and chooses the compa-
rability between two pictures by retribution the closeness of
these variety of areas.
1.4 Related work
This section presents state of the art developments carried
out in this domain over recent past years. Cimino [1] pro-
posed a hereditary span neural network architecture utiliz-
ing the dividing estimation of the data points on computa-
tional space. Froelich [2] proposed a definite time arrange-
ment demonstrating technique that utilizes data granules as
expected. Hmouz [3] proposed a time series oriented ex-
pectation model utilizing granular time series. Zhu [4] pro-
posed a hybrid variety of fuzzy model that consolidates the
T akagi–Sugeno (TS) fuzzy logic model and data division as
well as allocation technique. Musaylh [5] proposed a kind
of model that predicts transient power interest in Queens-
land utilizing SVR and ARIMA. Likewise, grouping tech-
nique for information investigation had been the subject of
various examinations, among them, a specific interest was
paid to setting based fuzzy C-implies (CFCM) grouping
utilizing fuzzy C-implies (FCM) grouping. Emmanuel D
[6] given description of Evolutionary Deep Networks for
Ef ficient Machine Learning. A. Sellami et al. [7] given
the similar investigation of dimensionality reduction strate-
gies for images examination and exploration. M Kayed et
al., in their work [8], performed categorization of the gar -
ments from fashion MNIST dataset exploiting CNN LeNet-
5 specific architecture. Image object’ s pattern analysis is
performed by Imran K. et al. [9]. Rim Rekik et al. [10]
proposed a computational cycle of gathering and separat-
ing information (measures including sites) from a rundown
of studies.
M. Stamenovic et al. [1 1] given a visual classifier , valuable
for deducing a record’ s general appearance, and a text clas-
sifier , for settling on content-informed decision choices.
Reshma P .K. et al. [12] given a soft computing frame-
work for web mining of multimedia data. MJ Sindhu [13]
given a framework for multimedia retrieval using web min-
ing. Seema S [14] given an extensive survey machine learn-
ing techniques for data mining. The measurable data and
probabilistic information is utilized for meta information
creation. In [15], Bayes’ hypothesis is exploited with ba-
sic freedom guesses among highlights. The augmentation
of data set application to deal with sight and sound arti-
cles requires synchronization of various media information
streams [16]. Peter et al. [17] portray the significance of
pre-preparing information in the web use mining and the
ef fect of errors to the investigation of the information dur -
ing this stage. It is fundamental for not disparage the in-
formation pre-processing stage during the time spent web
use mining, as the pre-handling stage straightforwardly in-
fluences the nature of the gained information. In [18], The
Remote Sensing (RS) image recovery strategies and appli-
cations are completely inspected. The assessment datasets
and measurements of RS image recovery are summed up.
Exhibitions on two sorts of exemplary RS image recovery
assignments are likewise talked about by the authors. G.
Suseendran et al.[35], in their work, have illustrated deep
learning Semi-structured T ree Miner for Data Stream’ s al-
gorithm focused on frequent pattern mining used in data
streams on semi-structured data. Singh et al., in their works
[36][37][38][39] have given robust frameworks and meth-
ods for processing multimedia data exploiting variety of
soft computing based computational primitives.
T able 1: Research gap / limitations
Appr oach Resear ch Gap
Froelich et al. [2] Increased complexity , Dif ficulty in capturing long-
term trends, Data sparsity .
Zhu et al. [4] Interpretability , Overfitting, Scalability , Lack of gen-
erality .
Musaylh et al. [5] Limited ability to capture complex patterns, Sensitiv-
ity to outliers, Limited ability to handle seasonality .
Sellami et al. [7] Loss of information, Selection of features to reduce,
Lack of transparency , Suf fers from Overfitting.
Kayed et al. [8] Computationally intensive, Lack of generalization,
Data requirements.
Peter et al. [17] Massive Data requirements, Computationally inten-
sive, Sensitivity to image quality .
Y ansheng et al. [18] Atmospheric interference more, Limited ground truth
data, Limited spectral coverage.
Some significant state of the art approaches along with
their limitations in the form of research gaps are given in
T able 1.4.
1.5 Contribution(s) in this paper
The contribution(s) in this work are as follows:-
– As research contribution, we proposed a pipelined
deep learning oriented methodology framework for
multimedia web-data mining based on extracted fea-
ture maps in planner projection as input.
– In order to prove the computational ef ficiency and to
validate its statistical behaviour , we have also pre-
sented the experimental evaluation on an standard
multimedia dataset. The obtained performance results
as compared to others significant existing approaches
in terms of various statistical measures/parameters.
1.6 Organization of the paper
The remaining portions of this paper are structured in a log-
ical and or ganized manner to present the research method-
ology and results in a clear and concise manner . In Sec-
tion 2, the proposed method is described in detail. This

84 Informatica 47 (2023) 81–90 M. Ravi et al.
section provides an overview of the key components and
techniques used in the research, highlighting the unique as-
pects of the proposed method. Section 3 is dedicated to
discussing the experimental evaluation and the results ob-
tained from the research. Finally , in Section 4, the pa-
per concludes with a conclusive summary of the research
and the future scope of this work. This section provides a
brief overview of the main contributions of the research and
highlights its potential impact on the field.
2 Pr oposed method
This section presents our proposed framework, architecture
and methodology overview along with the detailed algorith-
mic flow .
Algorithm 1 Feature Extraction
1: Five methods have been used in this phase -
{ – Color▶ performs color feature extraction
– Daisy▶ performs daisy features like intensity
– Edge▶ performs edge feature extraction
– Histogram of gradient▶ extracts gradient orien-
tation in the image
– Scale-invariant featur e transform (SIFT)▶ The
gradient at every pixel is viewed as an example of
a three-dimensional rudimentary component vec-
tor , shaped by the pixel area and the angle direc-
tion.
} 2: These methods individually does not provide the re-
sults as each method only deals with a particular spec-
ification like color , intensity , edges etc. So all these
methods are combined in the form of an Ensemble.
Algorithm 2 Inconsistency Measurement
1:∀Z = { V
1
,V
2
,··· V
n
} in approximation universal
space for particular conceptC , compute -
– ZC ={ x∈ U| [x]⊆ C} – ZC ={ x∈ U| [x]∩ C̸=∅} where, [x] is equivalence class
4: Compute,R
I
=ZC -ZC
5: IF (R
I
== NULL)
{ No inconsistent region
} Algorithm 3 Optimal V ariable’ s / Features Selection
1: BEGIN
2: F un (A, L, I,δ )
3: A: Set of attributes / features / variables
4: L: Label column
5: I: Set of instances
6: δ : Threshold
7: fun
cal()
: Function to calculate core features
8:R← fun
cal()
9: whileγ R
(L) <δ 10: I← I -POS
R
(L)
1 1: ∀ x∈ (A -R )
12: p
a
=|POS
R∪{ a} (L)| 13: q
a
←| EquivClass
MAXLEN
(POS
R∪{ a} (L))| 14: Choose a value with lar gest ofp
a
× q
a
15: R←R∪{ a} 16: ReturnR
17: END
2.1 Ar chitectur e and methodology overview
Computer vision techniques and deep learning algorithms
together are used for the feature extraction process. Each
image has been represented as a feature vector . Similarity
techniques such as Cosine Similarity and Euclidean Dis-
tance are used to measure the closeness between feature
vectors of the query image and the images available in the
dataset. Soft computing based statistical algorithmic mod-
ule is adopted for an optimized variable’ s selection.
2.2 Detailed algorithmic flow
Our approach Removes the typical method of storing pho-
tos in databases, which involves labelling them according
to their content and retrieving them using key words. In
this proposed methodology , multimedia web-data mining is
performed based on extracted feature maps in planner pro-
jection as input rather than based on the input meta-data
such as keywords, tags, or descriptions associated with the
image.
There are in total four main algorithmic modules in it. It is
primarily based on equivalence class evaluation pipelined
to feature maps onto planar projection. Algorithmic block
1 is performing features extraction. Algorithmic block 2 is
performing inconsistency measurement. Computationally
optimal variables’ selection is carried out in algorithm 3.
Algorithmic block 4 is performing transfer learning proce-
dure.

A New Multimedia W eb-Data Mining Approach… Informatica 47 (2023) 81–90 85
Algorithm 4 T ransfer Learning Proc()
1: VGGNET + RESNET ar chitectur e exploitation:
– Input→ fixed 224× 224 dimension pixels image
– Subtract mean RGB value, computed on the train-
ing set, from each pixel
– Use filters in convolution layer: 3× 3 dimension
– Fix convolution stride to 1 pixel; padding is 1
pixel for 3× 3 conv . layers
– Exploit Lr
MAXPOOLING(1)
··· Lr
MAXPOOLING(5)
– Max-pooling↬ { pixel window = 2× 2, stride =
2} – Process in Fully-connected layers: 4096 channels
– Process in soft-max layer
Our proposed methodology framework has some com-
putational advantages over other methods available in the
literature. The discussion regarding the same is given as
points below:-
– Proposed algorithmic module of Dimension reduction
based feature selection has the ability to reduce the
number of features or variables in a dataset, which
can make the dataset more manageable and reduce the
computational complexity of subsequent analyses or
modeling tasks. This can result in faster computation
and reduced resource requirements.
– Proposed module for transfer learning based classifi-
cation can learn complex non-linear relationships in
data without requiring explicit feature engineering,
which can save time and ef fort in the data mining pro-
cess.
– Our proposed methodology can perform well in tasks
with imbalanced datasets, where the number of exam-
ples in each class is dif ferent, by adjusting the decision
threshold. This can improve the model’ s performance
on the minority class.
3 Experimental analysis
This section presents the experimental evaluation in form of
simulation set-up details, dataset summary , exploited com-
putational libraries and packages, obtained results summary
and comparisons of our proposed method results with some
significant existing approaches.
3.1 Simulation set-up
In this study , the simulation environment is built up as fol-
lows: Our operating system was Ubuntu 18.04 L TS, and
our hardware included 8 GB of RAM and an Intel Core i7
4032U CPU processor running at 3.2GHz.
3.2 Dataset overview
For the purpose of experimental analysis, Corel 10K dataset
[34] is utilized here that features 10,000 pictures from vari-
ous substances such as dusk, seashore, bloom, building, car ,
horses, mountains, fish, food, and doorway , among others.
Each class includes 100 JPEG images with a resolution of
192times 128 or 128times 192. T en classes were chosen
for trial simulation purposes from the dataset’ s total of 100
classes.
3.3 Exploited computational libraries and
packages
Python 3.7.4 is used for the implementation. The compu-
tational libraries and packages used are - NumPy , SciPy ,
pandas, Matplotlib, Statsmodels and PyT orch. PyT orch is
an essential requirement for running ResNet and VGGNet
modules.
3.4 Results summary
The generated confusion matrix is a 10× 10 matrix as the
total number of classes considered here are 10. T able 2 rep-
resents the statistical performance measures corresponding
to each individual category in terms of precision, recall and
accuracy . Numerous test cases for experimental evaluation
are carried out and obtained results images are given for
input query instances and corresponding output result in-
stances (Figure 3 - Figure 12).
3.4.1 Confusion matrix
Figure 1: Confusion matrix

86 Informatica 47 (2023) 81–90 M. Ravi et al.
3.4.2 Statistical performance measur es
T able 2: Statistical performance measures
Label / Category Pr ecision Recall Accuracy
0 0.6 1 0.9
1 1 1 0.89
2 1 1 0.9
3 1 1 0.93
4 1 1 0.98
5 0.6 0.6 0.71
6 1 0.6 0.89
7 1 0.8 0.94
8 1 0.2 0.78
9 1 1 0.98
The depiction for statistical performance measures is rep-
resented as Figure 2.
Figure 2: Depiction for Statistical Performance Measures
Figure 3: i/p query instance (buildings)
Figure 4: o/p result instances (buildings)
Figure 5: i/p query instance (flowers)
Figure 6: o/p result instances (flowers)

A New Multimedia W eb-Data Mining Approach… Informatica 47 (2023) 81–90 87
Figure 7: i/p query instance (tr ees)
Figure 8: o/p result instances (tr ees)
Figure 9: i/p query instance (mountains)
Figure 10: o/p result instances (mountains)
Figure 1 1: i/p query instance (vegetables)

88 Informatica 47 (2023) 81–90 M. Ravi et al.
Figure 12: o/p result instances (vegetables)
3.5 Comparisons on performance metrics
In order to prove the computational novelty of proposed
framework, comparisons are performed with some signif-
icant existing approaches and the results are provided in ta-
ble 3. From this table, it can be observed that the proposed
framework outperforms over other existing approaches.
T able 3: Comparative analysis
Method Pr ecision V alues Recall V alues
MTH [30] 40.87 49.1
CPV -THF [31] 52.28 62.7
STH [32] 48.03 57.6
CMSD [33] 50.25 60.3
Proposed framework 92.0 82.0
4 Conclusion and futur e scope
The performance of media information mining is directly
influenced by the level of information contained in the data,
which can be recognized during the data pre-processing
stage. This paper presents a pipelined deep learning
methodology framework for multimedia web-data mining
based on extracted feature maps in planner projection as
input. The framework is designed to be oriented towards
deep learning techniques. An experimental evaluation is
also conducted on a standard multimedia dataset [34]. The
obtained performance results are compared with some sig-
nificant existing approaches in terms of various statistical
measures and parameters.
4.1 Futur e scope
The future scope of this research work is to analyze and
identify interrelationships within Multimedia data sets as
well as to derive a composite score from several dif ferent
sub-scores.
Refer ences
[1] Cimino, M.G.C.A.; Lazzerini, B.; Marcelloni, F .;
Pedrycz, W . Genetic interval neural networks for
granular data regression. Inf. Sci. 2014, 257, 313–330.
url: https://doi.or g/10.1016/j.ins.2012.12.049.
[2] Froelich, W .; Pedrycz, W . Fuzzy cognitive
maps in the modeling of granular time series.
Knowl.-Based Syst. 2017, 1 15, 1 10–122. url:
https://doi.or g/10.1016/j.knosys.2016.10.017.
[3] Hmouz, R.A.; Pedrycz, W .; Balamash, A. De-
scription and prediction of time series: A
general framework of granular computing. Ex-
pert Syst. Appl. 2015, 42, 4830–4839. doi:
https://doi.or g/10.1016/j.eswa.2015.01.060.
[4] Zhu, X.; Pedrycz, W .; Li, Z. A design of
granular T akagi-Sugeno fuzzy model through the
syner gy of fuzzy subspace clustering and opti-
mal allocation of information granularity . IEEE
T rans. Fuzzy Syst. 2018, 26, 2499–2509. doi:
https://doi.or g/10.1 109/tfuzz.2018.2813314.
[5] Musaylh, M.S.A.; Deo, R.C.; Adamowski, J.F .;
Li, Y . Short-term electricity demand forecasting
with MARS, SVR and ARIMA models using
aggregated demand data in Queensland, Aus-
tralia. Adv . Eng. Inform. 2018, 35, 1–16. doi:
https://doi.or g/10.1016/j.aei.2017.1 1.002.
[6] Emmanuel D, B.A. Bassett, EDEN: Evolutionary
Deep Networks for Ef ficient Machine Learning,
arXiv preprint doi arXiv: 1709.09161, 2017.

A New Multimedia W eb-Data Mining Approach… Informatica 47 (2023) 81–90 89
[7] A. Sellami and M. Farah, ”Comparative study of di-
mensionality reduction methods for remote sensing
images interpretation,” 2018 4th International Confer -
ence on Advanced T echnologies for Signal and Im-
age Processing (A TSIP), Sousse, 2018, pp. 1-6, doi:
10.1 109/A TSIP . 2018.8364490.
[8] M. Kayed, A. Anter and H. Mohamed, ”Clas-
sification of Garments from Fashion MNIST
Dataset Using CNN LeNet-5 Architecture,” 2020
International Conference on Innovative T rends
in Communication and Computer Engineering
(ITCE), Aswan, Egypt, 2020, pp. 238-243. doi:
https://doi.or g/10.1 109/itce48509.2020.9047776.
[9] Imran Khan, Asif Khan, Riaz Ahmed Shaikh, Ob-
ject analysis in image mining, 2015 2nd International
Conference on Computing for Sustainable Global De-
velopment (INDIACom), 1 1-13 March 2015, IEEE.
[10] Rim Rekik, Ilhem Kallel, Jor ge Casillas, Adel
M.Alimi, Assessing web sites quality: A systematic
literature review by text and association rules mining,
International Journal of Information Management,
V olume 38, Issue 1, February 2018, Pages 201-216.
doi: https://doi.or g/10.1016/j.ijinfomgt.2017.06.007.
[1 1] Marko Stamenovic, Sam Schick, Jiebo Luo, Machine
Identification of High Impact Research through T ext
and Image Analysis, 2017 IEEE Third International
Conference on Multimedia Big Data, 19-21 April
2017, IEEE, doi: 10.1 109/BigMM.2017.63.
[12] Reshma P .K., Lajish V .L., W eb Mining for Multime-
dia Data-A Soft Computing Framework, International
Journal of Scientific & Engineering Research, V olume
5, Issue 9, September -2014.
[13] Manda Jaya Sindhu, Y . Madhavi Latha, V . Samson
Deva Kumar , Suresh Angadi, Multimedia Retrieval
Using W eb Mining, International Journal of Recent
T echnology and Engineering (IJR TE) ISSN: 2277-
3878, V olume-2, Issue-1, March 2013.
[14] Seema Sharma, Jitendra Agrawal, Shikha Agar -
wal, Sanjeev Sharma, Machine Learning T ech-
niques for Data Mining: A Survey , 2013
IEEE International Conference on Computa-
tional Intelligence and Computing Research. doi:
https://doi.or g/10.1 109/iccic.2013.6724149.
[15] Luis Enrique Sucar , Bayesian Classifiers, Probabilis-
tic Graphical Models, Advances in Computer V i-
sion and Pattern Recognition pp. 41-62, (2015). doi:
https://doi.or g/10.1007/978-1-4471-6699-3_4.
[16] P . K. Y adav and S. Rizvi, ”An exhaustive study
on data mining techniques in mining of Multime-
dia database,” 2014 International Conference on Is-
sues and Challenges in Intelligent Computing T ech-
niques (ICICT), Ghaziabad, 2014, pp. 541-545, doi:
10.1 109/ICICICT .2014.6781339.
[17] Peter Svec, Lubomir Benko, Miroslav Kadlecik,
Jan Kratochvil, Michal Munk, W eb Usage Min-
ing: Data Pre-processing Impact on Found Knowl-
edge in Predictive Modelling, Procedia Computer
Science, V olume 171, 2020, Pages 168-178. doi:
https://doi.or g/10.1016/j.procs.2020.04.018.
[18] Y ansheng Li, Jiayi Ma, Y ongjun Zhang, Image re-
trieval from remote sensing big data: A survey , Infor -
mation Fusion, V olume 67, 2021, Pages 94-1 15. doi:
https://doi.or g/10.1016/j.inf fus.2020.10.008.
[19] Z. Pawlak, Information systems theoretical foun-
dations, Information systems V ol.6 (3):pp. 205-
218 (1981). doi: https://doi.or g/10.1016/0306-
4379(81)90023-5.
[20] Cunningham P ., Dimension reduction. T echnical Re-
port: UCD-CSI-2007-7, (2007).
[21] Y an, J., Zhang, B., Liu, N., Y an, S., Cheng, Q., Fan,
W ., Y ang, Q., Xi, W ., Chen, Z. Ef fective and ef ficient
dimensionality reduction for lar ge-scale and stream-
ing data preprocessing, IEEE T ransaction on Knowl-
edge and Data Engineering, V ol. 18, No. 3, 320-333,
(2006). doi: https://doi.or g/10.1 109/tkde.2006.45.
[22] J.-H. Seok, J.H. Kim, Scene text recognition
using a Hough forest implicit shape model
and semi-Markov conditional random fields,
Pattern Recogn. 48 (2015) 3584-3599. doi:
https://doi.or g/10.1016/j.patcog.2015.05.004.
[23] I.H.W itten, E. Frank, Data mining: Practicalmachine
learning tools and techniques, Mor gan Kaufmann,
2005.
[24] P . Shivakumara, T .Q. Phan, C.L. T an, A ro-
bust wavelet transform based technique for
video text detection, 2009 1285–1289. doi:
https://doi.or g/10.1 109/icdar .2009.83.
[25] C.H. Hansen, Fundamentals of acoustics, in: B.H.
Goelzer , C.H. Hansen, G.A. Sehrndt (Eds.), Occupa-
tional Exposure to Noise: Evaluation, Prevention and
Control,W orld Health Or ganization, Geneva, 2001.
[26] E. Uzun, H.T . Sencar , A preliminary examina-
tion technique for audio evidence to distinguish
speech from non-speech using objective speech qual-
ity measures, Speech Comm. 61-62 (2014) 1–16. doi:
https://doi.or g/10.1016/j.specom.2014.03.003.
[27] R. Bellman, R.E. Bellman, R.E. Bellman, R.E. Bell-
man, Adaptive control Processes: a guided tour ,
Princeton University Press, Princeton, 1961. doi:
https://doi.or g/10.1002/nav .3800080314.
[28] M. V erleysen, D. François, The curse of dimen-
sionality in data mining and time series pre-
diction, Computational Intelligence and Bioin-
spired Systems, Springer 2005, pp. 758–770. doi:
https://doi.or g/10.1007/1 1494669_93.

90 Informatica 47 (2023) 81–90 M. Ravi et al.
[29] C.M. Bishop, Pattern recognition and machine learn-
ing, Springer , 2006.
[30] G.-H. Liu, L. Zhang, Y .-K. Hou, Z.-Y . Li,
and J.-Y . Y ang, “Image retrieval based on
multi-texton histogram,” Pattern Recogni-
tion, vol. 43, no. 7, pp. 2380–2389, 2010. doi:
https://doi.or g/10.1016/j.patcog.2010.02.012.
[31] A. Raza, H. Dawood, H. Dawood, S. Shabbir , R.
Mehboob, and A. Banjar , “Correlated primary visual
texton histogram features for content base image re-
trieval,” IEEE Access, vol. 6, pp. 46595–46616, 2018.
doi: https://doi.or g/10.1 109/access.2018.2866091.
[32] A. Raza, T . Nawaz, H. Dawood, and H. Da-
wood, “Square texton histogram features for im-
age retrieval,” Multimedia T ools and Applica-
tions, vol. 78, no. 3, pp. 2719-2746, 2019. doi:
https://doi.or g/10.1007/s1 1042-018-5795-x.
[33] H. Dawood, M. H. Alkinani, A. Raza, H. Da-
wood, R. Mehboob, and S. Shabbir , “Correlated mi-
crostructure descriptor for image retrieval,” IEEE
Access, vol. 7, pp. 55206–55228, 2019. doi:
https://doi.or g/10.1 109/access.2019.291 1954.
[34] https://sites.google.com/site/dctresearch/Home/
[35] G. Suseendran, D. Balaganesh, D. Akila and
S. Pal, ”Deep learning frequent pattern mining
on static semi structured data streams for im-
proving fast speed and complex data streams,”
2021 7th International Conference on Optimiza-
tion and Applications (ICOA), 2021, pp. 1-8, doi:
10.1 109/ICOA51614.2021.9442621.
[36] Arappal, N., Singh, A., Saidulu, D. (2022, Septem-
ber). A Soft Computing Based Approach for Pixel
Labelling on 2D Images Using Fine T uned R-
CNN. In International Conference on Innovations
in Computer Science and Engineering (pp. 415-
424). Singapore: Springer Nature Singapore. doi:
https://doi.or g/10.1007/978-981-19-7455-7_61.
[37] Singh, A., T iwari, V ., T entu, A. N. (2022, March).
Ceiling improvement on breast cancer prediction
accuracy using unary KNN and binary LightGBM
stacked ensemble learning. In Proceedings of the Sev-
enth International Conference on Mathematics and
Computing: ICMC 2021 (pp. 451-471). Singapore:
Springer Singapore. doi: https://doi.or g/10.1007/978-
981-16-6890-6_34.
[38] Singh, A., T iwari, V . (2019). An optimal di-
mension reduction-based feature selection and
classification strategy for geospatial imagery . In-
ternational Journal of Knowledge Engineering
and Soft Data Paradigms, 6(2), 120-138. doi:
https://doi.or g/10.1504/ijkesdp.2019.102851.
[39] Singh, A., T iwari, V ., Gar g, P ., T entu, A. N. (2019).
Reasoning for Uncertainty and Rough Set-Based Ap-
proach for an Ef ficient Biometric Identification: An
Application Scenario. In Computational Intelligence:
Theories, Applications and Future Directions-V olume
II: ICCI-2017 (pp. 465-476). Springer Singapore. doi:
https://doi.or g/10.1007/978-981-13-1 135-2_35.