Wireless Personal Communications (2020) 112:169–192

https://doi.org/10.1007/s11277-019-07021-6

Color Object Detection Based Image Retrieval Using ROI

Segmentation with Multi‑Feature Method

Rohit Raja1 · Sandeep Kumar2 · Md Rashid Mahmood3

Published online: 1 January 2020

© Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract
Nowadays content-based image retrieval (CBIR) framework is drawing in consideration of
numerous analysts because of far-reaching applications found in numerous territories. In
this paper, a new CBIR methodology is proposed and adequacy of any CBIR framework
relies upon the features extracted from a color picture. In this work, firstly find the region
of interest of the image using Sobel and Canny method and later on output is applied on
HSV color space, it is clear to human vision eye. For classification, neural network is used
and categorized the data with class labels. The similarity distance is estimated between the
query image and stored image with different similarity metrics like Manhattan distance,
Euclidean distance, Chebyshev, Hamming distance and Jaccard distance. The experimental
result is estimated on accuracy, precision. The experiment performed on two well-known
databases i.e.: Corel-1k and Corel-5k dataset and new methodology proves the better accu-
racy results up to 87.33% and 68.93% respectively and improves the precision results also
up to 86.36% and 68.47% respectively. In this paper, results are also extended up to 80%.

Keywords CBIR · Sobel edge detection · Canny edge detection · HSV histogram · NN

1 Introduction

CBIR is a strategy to find and furthermore record pictures in the enormous arrangement
of database dependent on its visual substance, for instance, the premise of surfaces, hues,
shapes or spatial designs in its place of applying labels or different illustrative meta-
data catchphrases that may connect with pictures in the database. In conventional CBIR

* Rohit Raja
[email protected]
Sandeep Kumar
[email protected]
Md Rashid Mahmood
[email protected]
1
CSE Department, Sreyas Institute of Engineering and Technology, Hyderabad, India
2
ECE Department, Sreyas Institute of Engineering and Technology, Hyderabad, India
3
ECE Department, Guru Nanak Institutions Technical Campus, Hyderabad, India

13
Vol.:(0123456789)
170 R. Raja et al.

frameworks work is done through extricating one or additional multi-dimensional vectors

from all picture in the database, this methodology is finished in a back advance to begin
recovering. At question time, the couple of vectors are commonly separated from inquiry
picture and a likeness based capacity is utilized at that point to gauge the amount of fluc-
tuation between various pictures and inquiry picture vector in the database. Those pictures
have a similitude vector to inquiry vectors are the arrangement of last recovered result [1].
In this paper, a similarly little feature database is created from the underlying exten-
sive picture database. For this, a component vector comparing to every one of the pictures
in the substantial database is shaped utilizing the visual substance of that picture. This is
finished utilizing the extraction of shading and edge alongside other better subtleties by
discrete wavelet transform (DWT). Color is a standout amongst the most imperative visual
highlights for people. For the extraction of data identified with a color, make utilization of
four shading models in this paper, in particular, RGB, HSV, YCbCr and HSI. Color space
comprises of 3D spaces and color is utilized as a vector in it. Medicinal looks into demon-
strating that the human eye is more delicate to change in splendor than a change in color.
So transformation from RGB to HSI has improved the situation better outcomes [2].
The distinction among HSI and RGB is that HSI speaks to shading as splendor and two
shading contrast signals. In HSI, H is the tint, S is the immersion and depicts flags how
much shading is contaminated with white box and I signified the power. The HSI shading
space is an essential and alluring shading model for picture handling as it speaks to hues
similarly as the human eye detects hues. HSI shading model speaks to each shading with
three segments; tint, immersion, and power. For extricating the edges, Canny edge identi-
fier and Prewitt edge finders were utilized. The Canny edge locator is an ideal edge indica-
tor. It gives a low mistake rate as it has a reaction to edges and no reaction to non-edges.
The Prewitt edge channel is utilized to recognize edges by applying even and vertical chan-
nels in groupings. Here we consolidate both these channels to acquire better results [2].

2 Literature Survey

Kaur et al. [5] has presented the efficient approach through images are retrieve by enhanc-
ing the CBIR for feature extraction. By using these features, different classifier used i.e.
ANN, Bayes and Fuzzy Neural Network for classification and achieved good precision
results respectively 76%, 66% and 43%.
Varish et al. [6] has introduced a CBIR methodology based on color features compo-
nent. Author gave the similar importance to all three fundamental colors while evaluating
the images. Processing cost of proposed methodology is low as compared to others while
testing the standard datasets and achieved 84.0% results.
Zhao et al. [7] have presented a novel approach means Multi-Trend Structure Descrip-
tor (MTSD) for feature extraction. With the MTSD algorithm, both Low level features and
local spatial structure information are characterized easily. Authors evaluate this algorithm
on standard datasets and evaluated results demonstrate the satisfactory output i.e. 79.28%
and 62.98 respectively Corel-1k and Corel-5k datasets. Execution time of proposed meth-
odology for feature extraction is also satisfactory approximately 9.1615 s.
Zheng et al. [8] has introduced a fast image retrieval algorithm with a good efficiency.
This algorithm applied on standard datasets Corel-1k and Corel-5k respectively achieved
69.32% and 51.84% results while evaluation.

13
Color Object Detection Based Image Retrieval Using ROI… 171

Mikolajczyk et al. [9] has presented a interest region detector based descriptor approach.
While evaluation, extension of SIFT descriptor used and it shows the better performance
74.67% and 57.92% among the low dimensional descriptor on Corel-1k and Corel-5k data-
sets respectively.
Amjad et al. [10] has presented Convolution Neural Network (CNN) for feature extrac-
tion from pictures in CBIR framework. ED is utilized for the relationship among question
and put away pictures utilizing the removed highlights. Execution of the proposed work
is assessed utilizing exactness. This work demonstrates enhanced outcomes when con-
trasted with the current works and results achieved 95%, 97%, and 88% respectively Corel,
Caltech, and Li datasets.
Aasia et al. [11] has introduced SIFT for feature extraction, which essentially gives us
the key point in a picture. Filter picture feature calculation give a lot of pictures includes
that are not important so we utilize the improvement method BFOA (Bacteria foraging
optimization algorithm) to decrease the multifaceted nature, cost, vitality and Time utiliza-
tion. Finally proposed methodology improves the accuracy on animal, butterfly, facial and
flower images respectively 60%, 76.7%, 97%, and 98%.
Ekta et al. [12] have introduced a structure for picture recovery that applies different
nearby surface descriptors on Log-Gabor channels reaction. Tests have been directed on
standard Wang, VisTex, and OT-Scene databases and respectively recovery execution of
proposed structure percentage improved up to 9.06, 9.20 and 8.23.
Pradnya et al. [13] examined the idea and extent of substance IR framework. It likewise
incorporates the diagram of MPEG-7 edge histogram descriptor (EDH) to remove the sub-
stance from pictures.
Benavides et al. [14] displayed the CBIR method considering as descriptors the mean,
the standard deviation, and the homogeneity of every one of the few picture windows
exposed to examination, that is, every window goes about as a neighborhood picture dis-
trict exposed to the face investigation having a face focal point at its center.

3 Proposed Work

In this paper new methodology has been proposed with a combination of all three features
i.e. texture, color, and shape. The experiment performed on the two standard databases and
achieved better precision results. This algorithm consists of the following steps:

1. Image Acquisition First step of the new algorithm is to consider query image and stored
image as shown in Fig. 2.
2. Pre Processing In this step, resized the query image and stored image with 192 * 128
sizes.
3. ROI Segmentation In the present work for segmentation of region of interest different
edge detector method are used like Sobel and Canny as shown in Fig. 1.
4. Feature Extraction The HSV color space is progressively appropriate for human eye
recognition. The human eye can’t recognize an expansive number of hues in the mean-
time; shading quantization is utilized in this strategy.

Color quantization is to appoint a specific gathering of hues to demonstrate the picture

with the most extreme valuable data. Split the image into three components H, S, and V
and assign 8, 8, 8 bins to each component, thus resulting in 8 * 8 * 8 = 512. Edge orientation

13
172 R. Raja et al.

Fig. 1  a Resized query image

(192 * 128) and b After applying
the edge detector

Fig. 2  Flow chart of proposed

system Start

Query Image
Image Collection

Resize Image with

192*128

ROI Segmentation Using Sobel and

Canny Edge Detection

Features of Query Image (Color Coding, HSV

Histogram, Edge Histogram and Intensity
Information)

Similarity Measure (L1, Euclidean, Chebyshev,

Hamming and Jaccard)

Classify Data using Feed

Forward Neural Network

Retrieved
Image

Stop

13
Color Object Detection Based Image Retrieval Using ROI… 173

is used for shape features and it identifies the object edges and image information. Convert
the RGB image into HSV color space. Apply the Canny edge detector to extract edges
of an image. Let the H and V edge detector to it and store the resulting matrix as H1 and
V1. Now apply Sobel edge detection on the H2 and V2 matrix. Now combine the edges
obtained the matrix H and V and after that combine with the unmodified S to get HSV
image. Now this HSV image reverts to RGB image. Then find the histogram of R, G, B
matrices of this image. These histograms i.e. HR, HG, and HB will contain 256 bins as
shown in Fig. 3.
The similarity distance is estimated between the query image and stored image with dif-
ferent similarity metrics i.e. Manhattan Distance (MD), Euclidean Distance (ED), Cheby-
shev, Hamming Distance (HD) and Jaccard Distance (JD).
DM = |x1 − x2| + |y1 − y2| (1)
√
DE = (x1 − x2)2 + (y1 − y2)2 (2)

DCheb = max(|x1 − x2|, |y1 − y2|) (3)

DH = min{D(x, y) ∶ x, y ∈ C, x ≠ y} (4)

DJ = 1 − (|x ∩ y|∕ |x ∪ y|) (5)

5. Classification In this step, read the stored database and query image features. There
are mainly two standard databases: Corel-1k and Corel-5k dataset. In Corel-1k data-
set: Africa, Beach, Monuments, Elephant, Horses, Building, Food, Flower, Mountain,
and Dinosaur etc. and Corel-5k: Churches, Polar Regions, domestic animals, Lions,
Tiger, Costumed People, desert, Models etc. The learning process is done using the NN
method. Training set includes all the images from the image database and query picture
contains the subtitle. At that point contrast input inquiry picture subtitle and each class
mark in the Corel database. The query picture will be classified with a specific class
mark or not without much of stretching. In the event that the information query picture
has a place with any of the class names in the database, presently we can choose just
that coordinated class pictures in the database and perform recovery activities on that
chose a specific piece of the database dependent on the similitude coordinating.

13
174 R. Raja et al.

13
Color Object Detection Based Image Retrieval Using ROI… 175

4 Result Analysis

The experiments were performed on Corel-1k and Corel-5k standard database to evaluate
the proposed methodology. On Corel-1k dataset, every class comprises 1000 pictures of
size 192 × 128 in JPG organize. On Corel-5k dataset, every class comprises 5000 pictures
of size 192 × 128 in JPG design. In proposed methodology, the quantity of coordinated pic-
tures is differing from 1 to 20. Utilizing Sobel and Canny mix, we discover the article
edges which were not recognized appropriately in past work. The proposed methodology
retrieve the similar kind of image and output is measured in the form of accuracy, recall,
execution time and precision.

13
176 R. Raja et al.

Accuracy = 100 × ((TP + TN)∕N) (6)

Recall = TP∕(TP + FP) (7)

Precision = TP∕(TP + FN) (8)

TP: True Positive, TN: True Negative, FN: False Negative, FP: False Positive, N = Size
of dataset.
True positive means object is present and evaluation results detect the conditions.
True negative means object is not present and evaluation results doesn’t detect the
conditions.
False positive means object is not present and evaluation results detect the conditions
(Figs. 2, 3).
False negative means object is present and evaluation results doesn’t detect the
conditions.
In Figs. 4 and 5, it shows the standard datsets used for performing the experiments. In
Figs. 6, 7 and 8, it shows that edges of an object using Sobel and Canny edge detection
method. The output of proposed algorithms is shown in Figs. 9, 10, 11 and 12. Where
Fig. 9 shows the Retrieval of 20 Images for African People by Proposed Algorithm on the
standered dataset. In Fig. 10, the query image is bus and Retrieval of 20 Images for the
bus by Proposed Algorithm on the Corel-1k dataset. Retrieval of 20 Images for Playing
Cards by Proposed Algorithm on the Corel-1k dataset is shown in Fig. 13. Figures 14, 15
and 16 represents the Retrieval of 20 Images for Royal Cars, crockery, cards, argricultural
vechile by Proposed Algorithm on the Corel-5k dataset. Figures 17, 18 and 19 shows the
Accuracy, Precision, Feature Vector and feacture extraction comparision between MTSD
and Proposed method. Figures 20, 21 and 22 shows comparision of average presision,
Average accuracy, average retrival presicion value using different method MTSD and Pro-
posed method. Figure 23 Show Accuracy Comparison between various distance method on
Coreal-5k Dataset, Fig. 24 shows Comparision of Average Precision Value using a different

Fig. 3  Results on beverages image with modified edge detection

13
Color Object Detection Based Image Retrieval Using ROI… 177

Input Image
(RGB)

H S V

H1 (Canny) H2 (Sobel) V1 (Canny) V2 (Sobel)

Combine H, S & V and Make RGB

Y Cb Cr

Histogram

Hy Hcb Hcr

3-L DWT 2-L DWT 3-L DWT

Concatenation and Consider Intensity Information (HSI)

Feature Extraction

Fig. 4  Flow chart of proposed feature extraction methodology

method, Fig. 25 Shows the ROC by Proposed Algorithm on the Corel-1k and Corel-5k
dataset. Figure 26 shows Confusion Matrix by Proposed Algorithm on the Corel-1k data-
sets amd Corel-5k dataset. Figure 27 represpents MTSD Accuracy on Different Dataset
and Figure 28 represents Quantization level of intensity is fixed to 20, Accuracy Com-
parison between MTSD and Proposed method. Average accuracy and precision on Num-
ber of retrieval Images is represented in Table 1. The dimension of the feature vector and
feature extraction time was represented in Table 2. Table 3 represents the average value of

13
178 R. Raja et al.

Fig. 5  Experimental dataset of Corel-1k

Fig. 6  Experimental dataset of Corel-5k

13
Color Object Detection Based Image Retrieval Using ROI… 179

Fig. 7  Results on beverages image with edge detection

Fig. 8  Results on flags image with edge detection

13
180 R. Raja et al.

Fig. 9  Results on monuments image with edge detection

Fig. 10  Results on dinosaurs image with edge detection

13
Color Object Detection Based Image Retrieval Using ROI… 181

Fig. 11  Retrieval of 20 images for african people by proposed algorithm on the Corel-1k dataset

Fig. 12  Retrieval of 20 images for the bus by proposed algorithm on the Corel-1k dataset

13
182 R. Raja et al.

Fig. 13  Retrieval of 20 images for playing cards by proposed algorithm on the Corel-1k dataset

Fig. 14  Retrieval of 20 images for royal cars by proposed algorithm on the Corel-5k dataset

13
Color Object Detection Based Image Retrieval Using ROI… 183

Fig. 15  Retrieval of 20 images for crockery by proposed algorithm on the Corel-5k dataset

Fig. 16  Retrieval of 20 images for agricultural vehicle proposed algorithm on the Corel-5k dataset

13
184 R. Raja et al.

Accuracy Comparision
90
80
70
60
Percentage

50
40 Corel-1k
30
Corel-5k
20
10
0
MTSD Proposed
Database

Fig. 17  Shows accuracy comparison between MTSD and proposed method

Precision Comparision
100
90
80
70
Percentage

60
50
Corel-1K
40
30 Corel-5K
20
10
0
MTSD Proposed
Database

Fig. 18  Shows precision comparison between MTSD and proposed method

Comparision of Feature Vector

and Feature Extraction Time
250
Dimension Value and Seconds

200

150
Dimension
100 Time

50

0
MTSD Proposed

Fig. 19  Shows comparison between MTSD and proposed method

13
Color Object Detection Based Image Retrieval Using ROI… 185

Precision Comparision
100

80
Percentage
60

40 Corel-1k
Corel-5k
20

0
MTSD Proposed
Database

Fig. 20  Shows comparision of average precision value using a different method MTSD and proposed
method

Accuracy Comparision
100
90
80
70
Percentage

60
50
40 Corel-1K
30 Corel-5K
20
10
0
MTSD Proposed
Database

Fig. 21  Shows comparision of average accuracy value using a different method

MTSD precision & propoed prescision and MTSD accuracy and proposed accuracy when
the quantization level of color is 108. Table 4 the average retrieval precision of Proposed
Method on different similarity metrics. Table 5 the average retrieval precision of proposed
method on different similarity metrics. Table 6 represents average precision on number of
retrieval Image.

13
186 R. Raja et al.

Precision Comparison among Distances on

Corel-1k dataset
100

90

80

70

60 L1
Percentage

Euclidean
50
Chebyshev

40 Hamming
Jaccard
30

20

10

0
African Beach Monuments Elephant Flower

Fig. 22  Shows accuracy comparison between various distance method on Coreal-1k dataset

Precision Comparison among Distances on Corel-

5k dataset
80

70

60

50
L1
Percentage

Euclidean
40
Chebyshev
30 Hamming
Jaccard
20

10

0
Traffic Signals Playing Cards Flag Agricultural Trees
Vehicle

Fig. 23  Shows accuracy comparison between various distance method on Coreal-5k dataset

13
Color Object Detection Based Image Retrieval Using ROI… 187

AVERAGE PRECISION ON NUMBER OF

RETRIEVAL IMAGE
100
90
80
70
Percentage

60
50
40
30 Corel-1k
20 Corel-5k
10
0

Fig. 24  Shows comparision of average precision value using a different method

Fig. 25  ROC by proposed algorithm on the Corel-1k and Corel-5k dataset

13
188 R. Raja et al.

Fig. 26  Confusion matrix by proposed algorithm on the Corel-1k datasets amd Corel-5k dataset

Fig. 27  MTSD accuracy on dif- MTSD Accuracy (%)

ferent dataset
0%

Corel-5k
50.98% Corel-1k Corel-1k
55% Corel-5k

Fig. 28  Quantization level of Accuracy Comparison

intensity is fixed to 20, accuracy
100% 86.66 70.53
comparison between MTSD and
90%
proposed method
80%
70%
60% Proposed
50% Accuracy (%)
40% 64.86 52.97
MTSD Accuracy
30%
20% (%)
10%
0% 0
Corel-1k Corel-5k

13
Color Object Detection Based Image Retrieval Using ROI… 189

Table 1  Average accuracy and Dataset MTSD accu- Proposed MTSD pre- Proposed
precision on number of retrieval racy (%) accuracy (%) cision (%) precision
images (%)

Corel-1k 68.50 87.33 55.17 86.36

Corel-5k 52.42 68.93 53.44 68.47

Table 2  The dimension of Methods Dimension Feature

the feature vector and feature extraction
extraction time time (s)

MTSD 179 9.1615

Proposed 219 5.026

Table 3  The table represent the average value of MTSD precision and proposed precision and MTSD accu-
racy and proposed accuracy when the quantization level of color is 108
The quantization level of edge The quantization level of color is 108
orientation is 9
MTSD precision Proposed precision MTSD accuracy Proposed
(%) (%) (%) accuracy
(%)

Corel-1k 55 79.41 64.86 86.66

Corel-5k 50.98 69.82 52.97 70.53

Table 4  The average retrieval Category Proposed precision of similarity metrics (%) on Corel-
precision of proposed method on 1k dataset
different similarity metrics
L1 Euclidean Chebyshev Hamming Jaccard

African 79.66 69.81 90 88.88 81.66

Beach 85 87.27 79.36 92 82
Monuments 81.03 80.39 78.12 75 83.67
Elephant 70 71.42 90 91.93 84.21
Flower 89.13 85.50 84.31 89.47 77.58

Table 5  The average retrieval precision of proposed method on different similarity metrics
Category Proposed precision of similarity metrics (%) on Corel-5k dataset
L1 Euclidean Chebyshev Hamming Jaccard

Traffic signals 64.06 67.03 65.60 67.09 70.86

Playing cards 64.33 65.85 67.40 69.32 65.96
Flag 64.33 68.08 73.6 66.89 68.28
Agricultural vehicle 70.30 53.67 59.90 66.58 71.42
Trees 71.57 67.43 63.59 65.07 71.24

13
190 R. Raja et al.

Table 6  Average precision on number of retrieval image

Dataset Kaur [5] Varish [6] M. Zhao [7] L. Zheng [8] K. Mikolajczyk Proposed
precision precision (%) precision (%) precision (%) [9] precision (%) precision
(%) (%)

Corel-1k 73.4 80.4 79.28 69.32 74.67 86.36

Corel-5k – – 62.98 51.84 57.92 68.47

5 Conclusion

This paper has displayed that CBIR is utilized for looking and getting to the pictures from
large picture datasets. It recouped pictures based on picture descriptors. For color fea-
tures, we implement the color coding for finding color regions. In this work, firstly find
the region of interest (ROI) of the image using Sobel and Canny method. HSV histogram
is used for color descriptors. For finding edges used edge detector method, it detects an
object from an image and extracts features. This algorithm is better than the previous
algorithm. To detect the highest rank of an image used MD and HD similarity metrics
and it is far improved than other distance metrics. In the present work, the precision and
accuracy have reached up to 79.41% as well as 86.66% for Corel-1k dataset for N = 20 and
69.82% as well as 70.53% for Corel-5k dataset. In future work, the proposed method will
be tested on other database like Caltech database for improve object retrieval.

References
1. Arya, D., & Jha, J. (2016) Global and local descriptor for CBIR and image enhancement using multi-
feature fusion method. International Journal of Research—GRANTHAALAYAH.
2. Agarwal, S., Verma, A. K., & Dixit, N. (2011). Content-based image retrieval using color edge
detection and discrete wavelet transform. In International conference on issues and challenges in
intelligent computing techniques (ICICT), February 7–8, 2011 (pp. 368–372)
3. Zinzuvadia, K. M., Tanawala, B. A., & Brahmbhatt, K. N. (2015). A survey on feature-based image
retrieval using classification and relevance feedback techniques. IJIRCCE, 2, 1253.
4. Dixit, N., Tiwari, S. K., & Sharma, P. (2016). A new algorithm for CBIR using bi-cubic interpolation
with color coding and different level DWT. In 2016 IEEE (ICCUBEA) (pp. 1–6).
5. Kaur, M., & Dhingra, S. (2017). Comparative analysis of image classification techniques using statisti-
cal features in CBIR systems. In I-SMAC 2017 (pp. 265–270). IEEE.
6. Varnish, N., & Pal, A. K. (2015). Content-based image retrieval using statistical features of color
histogram. In 2015 3rd international conference on signal processing, communication and net-
working (ICSCN). IEEE
7. Zhao, M., Zhang, H., & Sun, J. (2016). A novel image retrieval method based on multi-trend structure
descriptor. Journal of Visual Communication and Image Representation, 38, 73–81.
8. Zheng, L., Wang, S., Liu, Z., & Tian, Q. (2015). Fast image retrieval: Query pruning and early ter-
mination. IEEE Transactions on Multimedia, 17(5), 648–659.
9. Mikolajczyk, K., & Schmid, C. (2005). A performance evaluation of local descriptors. IEEE Trans-
actions on Pattern Analysis and Machine Intelligence, 27(10), 1615–1630.
10. Shah, A., Naseeml, R., Sadia, S. I., & Shah, M. A. (2017). Improving CBIR accuracy using convo-
lutional neural network for feature extraction. IEEE 2017 (pp 1–5).
11. Ali, A., & Sharma, S. (2017). Content-based image retrieval using feature extraction with machine
learning. In ICICCS 2017 IEEE (pp 1048–1053).
12. Walia, E., & Verma, V. (2016). Boosting local texture descriptors with Log-Gabor filters response for
improved image retrieval (pp. 1–12). London: Springer. https​://doi.org/10.1007/s1373​5-016-0099-2.
13. Vikhar, P., & Karde, P. (2016). Improved CBIR system using edge histogram descriptor (EHD) and
support vector machine (SVM). IEEE (pp. 1–5).

13
Color Object Detection Based Image Retrieval Using ROI… 191

14. Benavides, C., Villegas, J., Member, IEEE, Román, G., & Avilés, C. (2016). Face classification by
local texture analysis through CBIR and SURF points. IEEE Latin America Transactions, 14(5),
2418–2434.
15. Raja, R., Sinha, T. S., Patra, R. K., & Tiwari, S. (2018). Physiological trait based biometrical authen-
tication of human-face using LGXP and ANN techniques. International Journal of Information and
Computer Security Special Issue on: “Multimedia Information Security Solutions on Social Networks,
10(2/3), 303–320.
16. Raja, R., Sinha, T. S., & Dubey, R. P. (2015). Recognition of human-face from side-view using pro-
gressive switching pattern and soft-computing technique. Association for the advancement of model-
ling and simulation techniques in enterprises, Advance B (Vol. 58(1), pp. 14–34), ISSN:-1240-4543.
17. Kumar, S., Singh, S., & Kumar, J. (2018). Live detection of face using machine learning with multi-
feature method. Wireless Personal Communication Springer Journal (SCI). https​://doi.org/10.1007/
s1127​7-018-5913-0.
18. Kumar, S., Singh, S., & Kumar, J. (2018). Automatic live facial expression detection using genetic
algorithm with haar wavelet features and SVM. Wireless Personal Communication Springer Journal
(SCI). https​://doi.org/10.1007/s1127​7-018-5923-y.

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Dr. Rohit Raja has received Ph.D. in Computer Science and Engineer-
ing form CVRAMAN University in 2016. His main research interest
includes Face recognition and Identification, Digital Image Processing,
Signal Processing and Networking. Presently he is working as Profes-
sor in CSE Department, Sreyas Institute of Engineering and Technol-
ogy, Hyderabad India. He has authored several Journal and Conference
Papers. He has a good Academics and Research experience in various
areas of Computer Science and Engineering. He has filed successfully
8 Patents. He has been received 2 times invitation being a Guest in
IEEE Conferences. He has published 75 research papers in various
International/National Journals (including IEEE, Springer etc.) and
Proceedings of the reputed International/National Conferences (includ-
ing Springer and IEEE). He has been nominated in the board of edi-
tors/reviewers of many peer-reviewed and refereed Journals (including
IEEE, Springer).

Dr. Sandeep Kumar is presently working as a Professor in the Depart-

ment of Electronics and Communication Engineering, Sreyas Institute
of Engineering and Technology, Hyderabad, India. He has a good Aca-
demics and Research experience in various areas of Electronics and
Communication. His area of research includes Embedded System,
Image processing, and Biometrics.. He has filed successfully 6 Patents.
He has been received 2 times invitation being a Guest in IEEE Confer-
ences. He has published 60 research papers in various International/
National Journals (including IEEE, Springer etc.) and Proceedings of
the reputed International/National Conferences (including Springer
and IEEE). He has been awarded by “Best Paper Presentation” in
Nepal and India respectively 2017 and 2018. He has been awarded for
“Best Performer Award” in Hyderabad, India, 2018. He has been
awarded also “Young Researcher Award” in Thailand, 2018. He is an
active member of 13 various Professional International Societies. He
has been nominated in the board of editors/reviewers of 23 peer-
reviewed and refereed Journals (including IEEE, Springer). He is also
attended 24 of seminars, workshops and short-term courses in IITs etc.

13
192 R. Raja et al.

Dr. Md Rashid Mahmood Graduated in Electronics and Communica-

tion Engineering from Jamia Millia Islamia (A Central University),
New Delhi in 2002, did his Master in Electronics and Communication
Engineering from Maharshi Dayanand University, Rohtak in 2009 and
Ph.D from Department of Electronic and Communication Engineer-
ing, Jamia Millia Islamia (A Central University), New Delhi in 2017.
He has authored and Co-authored more than thirty research papers in
International/National Journals and proceedings in Symposia. His
research area of interest includes signal processing, propagation and
Scattering of transmission lines, design and applications of microwave
filters and antennas.

13