Proceedings of the Third International Conference on Electronics and Sustainable Communication Systems (ICESC 2022)

IEEE Xplore Part Number: CFP22V66-ART; ISBN: 978-1-6654-7971-4

AI-SMLA: An Artificial Intelligence based

Smart Machine Learning Algorithm for
2022 3rd International Conference on Electronics and Sustainable Communication Systems (ICESC) | 978-1-6654-7971-4/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICESC54411.2022.9885374

Complex Image Segmentation Issues in

Vertex Image Processing
Dr. R. Pradeep Kumar Reddy, Assistant Dr. S. Kiran, Dr.A.Ashok
Professor, Associate Professor, Y.S.R Engineering Kumar,AssociateProfessor,Y.S.R
Y.S.R Engineering College of YVU, College of YVU,Proddatur, Andhra Engineering College of YVU,Proddatur,
Proddatur, Andhra Pradesh, India. Pradesh, India. Andhra Pradesh, India.
[email protected] [email protected] [email protected]

Abstract—Today's machine learning is considered as one of the processing.Digital filters are used to blur and sharpen digital
artificial intelligence technologies used in many ways. Its images [12]-[14]. Filtering in the location sphere can be done
functions are very accurate, from receiving the given input by rotating in the frequency field (Fourier) by covering
data to calculating, measuring and outputting it. It is used to specially designed cores (filter array) or certain frequency
calculate a valid result without user intervention. In that order bands. Dig ital cameras usually have special hardware for image
the applications of this machine learning method in digital
processing (special chips or circuits embedded in other chips)
image processing technology are constantly increasing. This
technology is used for everything from overcoming facial to convert source data from image sensors to a color-adjusted
identities to calculating the various photo identities of mobile image in the form of a standard image file.
devices currently in use. In this paper a smart machine An image processing refers to the work that is done not only on
learning based algorithm has been designed to effectively the image but also on the enclosure attributes. Image processing
handle vertex image processing environment and complex is a comb ination of changes in different data sets such as
image segmentation functions. Its special feature is that it brightness, sharpness, color, contrast, filter, resolution and size
examines the given photos based on its dead pixels and [15]-[16]. It is very helpful to analyze the information of those
calculates the correct results. There will be no human particular images and extract the required information. Due to
intervention in this system. The proposed smart algori thm was the high level o f develop ment in recent times, so me of the
compared with some existing algorithms.
problems that arise with this technology are likely to be high
Keywords: Artificial intelligence, accurate, digital image [17]. At present most of the businesses are doing special
processing, photo identities, mobile devices, machine learning, activities like d isplaying their products, highlighting their
vertex image processing, image segmentation. advertisements from other prostitutes, impressing the customers
through format authentication, and creating mu ltip le d isplays in
I. INTRODUCTION one frame through sorting processes [18]. All o f these are based
The special features of the machine learn ing system in on the digital system [19]. Its operation is highly comp lex and
general are an arbitrary art ificial intelligence system its accuracy is high when there are some efficient and
designed for individual use methods and to define the spontaneous machine learn ing methods to manage it [20]. The
advanced process modules of organizations [4]. The speed following volu mes on this paper detail the informat ion of
of data handling and the ability to calculate efficient results previous scientists on it, the proposed algorithm, the
in less time are the reasons why most people seek these. It is experiment and the results.
further enhanced by the accuracy of its results and its ability II. RELATED WORKS
to make meaningful decisions. Image processing methods in Abbas et al [1] introduced an improved image segmentation
general have co me to a standstill today. The do minance of system. It carried out analyzes based on classified image
image processing technology has increased everywhere, blocks. The significance of this was that its various image
fro m face scans to builder images on your s mart phone blocks were interconnected by its link codes which were
cameras [8]-[11]. Furthermore the series of images continuously divided into smaller wages based on its accuracy.
generated by the image processing system has a nu mber of The accuracy of those images was calcu lated based on these
functions that further enhance its accuracy. Each work lin ks. But only its coupling speed during grouping operations
package is involved in enhancing the data of specific images made connectivity more difficult.Berto zzi et al. [2] handled
while executing it. Doing so tends to increase its size and limited image sets to sound the map. The significance of this is
the ratio of data such as resolution. Thus the impact on it that the image blocks divided into categories have its
will g reatly increase with increasing work. For example, integration paths based on small groups. This method of
when an image is processed and its work is co mpleted and converting it into a graph and configuring it based on its
the size of that particular image increases, it may need to be analytical results.Merkurjev et al [3] proposed a method of
compressed to text it. Then when you compress them based splitting the existing frame modules into video mode and
on those image blocks it goes back to the old state. Current making changes to its imaging mode. Image definit ions give
technologies help to rectify minor problems in image good results in this way but publishing them together as a video

978-1-6654-7971-4/22/$31.00 ©2022 IEEE 1361

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on June 19,2025 at 10:22:08 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Electronics and Sustainable Communication Systems (ICESC 2022)
IEEE Xplore Part Number: CFP22V66-ART; ISBN: 978-1-6654-7971-4

does not achieve excellent accuracy. Van Gennip et al [5] 5. If (segment = ordered)
analyzed accurate maps defined on the basis of image 6. Then convolute the image
volume. In this method the analysis and segmentation 7. Then pooling the image
methods came out better and the image resolution that 8. Then enhance the attributes
integrates it was better. But due to the large size of the 9. Else go-to step 3
images, problems arose in co mpressing it.Cavalcanti et al. 10. Group the images
[6] proposed microscopic image block classification. In this, 17. end
special codes were given to the image divided into blocks
based on the dot pixel and grouping functions were created That is, the input information first divides the blocks into pixel
based on it. This process gave good results.Chan.T.F et al rows. Depending on the clarity of the given inputs divided into
[7] categorized image groups based on vectors. That is, each height * width * dimension.Based on this, the machine learn ing
separated image b lock is linked to each vector and then its algorith m that calculates this selects the correct Measurement
blocks are classified. But even though the accuracy of this method. This helps to give known aspect vectors.The proposed
muran was high, the time taken to link the images took a algorithm consists of two modules.
long time. And sometimes the images are inco mplete 3.1. CONVOLUTION LAYER:
vectors of connection where proper alignment of the pixels It helps to handle co mplex image blocks. The mat rix modules
is not possible. on which it is based help to identify the special features present
III. PROPOS ED METHOD in the image and ext ract the image module fro m the unwanted
The proposed Artificial intelligence based smart machine data. Image square blocks that are rando m in this layer set
learning algorithm (AI-SM LA) is demonstrated in fig 1. A combine with the dot filter to fu rther enhance its functionality.
specific pipeline is always created when machine learning The higher the rating of the filter and patch modules the higher
algorith ms are created. Its main purpose is that the inputs the output of the convolution module. Perhaps the amount of
given fro m that pipeline data take the learning steps required output available will be significantly lower when the
for machine learning methods. The currently proposed performance of either of these two decreases. The proposed
algorith m says the same thing. That is, the first proposed convolution layer operation is shown in figure 2.
algorith m fo r machine learning should always use high -
accuracy data to maximize its accuracy. Therefore its 4 3
planning is designed to make it work well. High resolution 6 2 2 4
8 4
should be ensured where images are obtained as input based 1 2 3 1
on computer vision. This whole algorith m works on its 2 1
4 6 8 0
basis. It can be used to add and remove high resolution 1 -
images. 4 6 2 2 1 2 0 8
0 2
4 1 * - =
Input Preproces S egmenta Convoluti 1 2 8 2 2 0
2 2 2
Image sing tion on -
2 2 1 1 2 0
2 8 2 Output
6 8 0 4
Image
3 2 2 1 2
8
8 2 6 0 4 Filter
Original Image
Pooling
Imag Image Attribute
e s Fig 2: Operation of proposed convolution layer
Grouping
O utp Enhance Upp Ave
ut
ment
er rage The image seg mentation up-to the ‘n’ th level image
Pool Pool segments are denoted as follows,
ing ing
( ) ∫ ( ) ∫ ( ( ))
Fig 1: Proposed system block diagram
(1)
First all the inputs are extracted and then their entire format Where,
is changed to the extent required by the machine. The IS (a) = convolution segment output;
carving method is imp lemented by applying the required = subnet of the given image
parts in the next given image. It does not know it as forces. a(p) = two-phase image resolution concentration
Here the matrix processes present in the blocks of numbers
form the basis of this carving process. 3.2. POOLING LAYER:
Algorith m : Art ificial intelligence based smart machine When detecting certain features of images using the
learning algorithm convolution layers mentioned above, those blocks will have
1. Start mu ltip le feature levels. These types of maps are created when
2. Enter the input images the convolution process is used between the inserted image and
3. Slice the image as per the dot matrix method the special filters. Thus the size of the image is likely to
4. Provide the input segment expand. This pooling process is used to reduce this. In this

978-1-6654-7971-4/22/$31.00 ©2022 IEEE 1362

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on June 19,2025 at 10:22:08 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Electronics and Sustainable Communication Systems (ICESC 2022)
IEEE Xplore Part Number: CFP22V66-ART; ISBN: 978-1-6654-7971-4

method the pixel values in each volu me of the image are Where,Gj is the total number of segmented input image
reduced to a maximu m. These increase the accuracy of the commands
workings of the machine learning system as they create
movement indiv idually for each set of inputs. The accuracy The table 1 presents the Measurement of image recognition
is further enhanced by the fact that their functions operate between existing ISM, DIM G, GM BO, MPSLSand proposed
with t wo separate modules.The proposed pooling layer AISMLA
operation is shown in figure 3. Table 1: Measurement of input image recognition
4 6 10 48 34 4 No.of Input image recognition in (% )
22 12 14 26 38 10 Samples ISM DIMG GMBO MPSLS AISMLA
26 8 8 2 2 1 26 48 100 75.18 81.51 80.15 79.27 94.75
1 42 2 12 8 2 42 24 200 75.51 83.01 80.74 81.14 95.76
26 28 10 2 8 14 300 76.85 84.12 81.72 81.97 95.92
38 22 26 8 10 24 400 77.99 84.5 82.93 82.88 96.88
500 79.04 85.51 84.07 83.8 96.45
12 18 600 79.75 86.44 85.18 85.13 97.65
20 10 700 81.05 87.44 85.88 86.21 97.81

Fig 3: Operation of proposed pooling layer for the discrete

4.2. M EAS UREMENT INPUT IMAGE REJECTION
operations, The proposed algorithm usually separates the image blocks
( ) ∑ ( ) (2) with different data given inside. Some p ixels in that category
Where, have very low resolution. You can skip this and design other
( ) = diagonal matrix of the given image parts of that particular image gallery as it will take more t ime to
upgrade. Thus its accuracy is not affected.
IV. RES ULTS AND DIS CUS S ION
The proposed Artificial intelligence based smart machine ()
( )
learning algorith m (AI-SM LA) was co mpared with the ( )
(4)
existing an imp roved segmentation method (ISM), The
Diffuse interface models on graphs (DIM G), Graph M BO The below table 2 presents the Measurement of picture block
method for standoff detection (GM BO) and Macroscopic denial between existing ISM, DIM G, GM BO, M PSLSand
pigmented skin lesion segmentation (MPSLS) proposed AISMLA
Table 2: Measurement of input image rejection
There are the following parameters are evaluate the water
No.of Input image rejection in (% )
quality. That is the Segmentation accuracy, input image
Samples ISM DIMG GMBO MPSLS AISMLA
recognition, input image rejection, Seg mentation precision,
100 24.82 18.49 19.85 20.73 5.25
Segmentation recall, and Seg mentation F1-Score. Befo re
understand the quality rate of the parameters, will know 200 24.49 16.99 19.26 18.86 4.24
about the following, 300 23.15 15.88 18.28 18.03 4.08
Positive-T (TP)–It’s the perfect predicted correct or above 400 22.01 15.5 17.07 17.12 3.12
thecalibration level. 500 20.96 14.49 15.93 16.2 3.55
Negative-T (TN)–It’s the negative prediction values below 600 20.25 13.56 14.82 14.87 2.35
the calibration level. 700 18.95 12.56 14.12 13.79 2.19
Positives-F (FP) – When the exact values are in calibration
level and the predicted samples are in same level 4.3. M EAS UREMENT OF SEGMENTATION ACCURACY :
Negative-F (FN) – When the exact values are in calibration The Seg mentation accuracy is the parameter which describes
level butthe predicted samples are in different level the ratio between completely predicted Seg mentation input
images fro m the given pieces to the total number of collected
4.1. M EAS UREMENTO F INPUT IMAGE image pieces. When the rate of Seg mentation accuracy is high
RECOGNITION then the given output image sample getting high quality rate.
Usually the data given in the inputs are very large volumes.
The basic data of the various image modules running on the Accuracy Measurement = (5)
basis of these modules must be largely formatted. Special The below table 3demonstrates the various measurement
features of this design will take into account the data from comparison of the Segmentation accuracy values between the
its various angles. The volumes of this data can be existing ISM, DIMG, GMBO, MPSLS and proposed AISMLA
calculated as follows. 4.4. M EAS UREMENT OF S EGMENTATION
Then, the un-segmented image recognition of a method is PREC IS ION : Seg mentation precision measurement is the ratio
given by
∑ between the positive true samples and total true samples. The
(3)
total true samples are calculated by the sum of positive true
samples and false positive samples.

978-1-6654-7971-4/22/$31.00 ©2022 IEEE 1363

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on June 19,2025 at 10:22:08 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Electronics and Sustainable Communication Systems (ICESC 2022)
IEEE Xplore Part Number: CFP22V66-ART; ISBN: 978-1-6654-7971-4

Precision Measurement = Table 5: Measurement of recall rate

(6) No.of Recall rate in (% )
Samples ISM DIMG GMBO MPSLS AISMLA
Table 3: Measurement of accuracy measurement
100 64.25 75.57 76.15 74.58 94.58
No.of Accuracy measurement in (% ) 200 65.74 77.54 78.57 76.78 94.57
Samp DIM GMB MPSL AIS 300 66.54 78.67 78.98 77.58 95.77
ISM
les G O S MLA
400 68.87 79.86 80.58 78.25 96.25
100 72.88 79.21 83.55 82.01 93.84 500 69.88 80.25 82.9 79.68 97.68
200 73.21 80.71 84.14 83.88 94.88 600 70.52 81.77 84.15 80.77 98.84
300 74.55 81.82 85.12 84.71 95.01 700 71.18 82.01 86.88 81.25 99.61
400 75.69 82.2 86.33 85.62 95.97
500 76.74 83.21 87.47 86.54 95.54 4.6. M EAS UREMENT OF SEGMENTATION F1-SCORE:
600 77.45 84.14 88.58 87.87 96.78 It’s measured by the average sample values of Seg mentation
700 78.75 85.14 89.28 88.74 96.89 precision and Segmentation recall of the samples.
( )
F1-Score Measurement = (8)
The below table 4 demonstrates the various measurement ( )
comparison of the Seg mentation precision values between
the existing ISM, DIM G, GM BO, MPSLS and proposed The below table 6 demonstrates the various measurement
AISMLA comparison of the Segmentation F1-Score values between the
Table 4: Measurement of precision measurement existing ISM, DIMG, GMBO, MPSLS and proposed AISMLA
Table 6: Measurement of F1-Score
No.of Precision measurement in (% ) No.of F1-Score in (% )
Sample DIM GMB MPSL AIS ML Samples ISM DIMG GMBO MPSLS AISMLA
ISM 100 72.76 71.94 73.64 70.38 94.74
s G O S A
74.1 200 72.65 71.96 73.47 70.11 94.24
100 71.47 75.99 73.57 94.58
4 300 72.63 72.84 74.2 70.41 94.36
75.7 400 75.73 75.67 77.54 73.92 97.59
200 73.21 77.57 74.99 95.87
7 500 76.93 76.99 78.27 75.24 97.97
76.2 600 77.54 77.82 79.16 75.78 98.54
300 75.55 79.77 76.25 96.88
5 700 77.95 78.22 79.24 76.08 98.24
77.5
400 76.36 81.4 78.24 97.77
4 4.7. M EAS UREMENT OF RECOGNITION DURATION :
79.6 The Measurement duration is nothing but the time taken to
500 78.65 82.54 80.71 98.14
5 calculate the prediction of two different images.
81.1 Recognition Duration = (9)
600 80.58 84.74 82.15 99.18
4
82.9 The below table 7 demonstrates the various measurement
700 82.31 85.89 83.87 99.95 comparison of the Segmentation recognition duration between
5
the existing ISM, DIM G, GM BO, M PSLS and proposed
4.5. M EAS UREMENT OF SEGMENTATION RECALL:
AISMLA
Segmentation recall measurement is the rat io between the
Table 7: Measurement of recognition duration
positive true samples and the sum of positive true samples
No.of Recognition Duration in (ms)
and false negative true samples.
Samples ISM DIMG GMBO MPSLS AISMLA
Recall Measurement =
2000 3510 3844 2617 2771 1214
3000 4287 4401 3022 3155 1380
(7) 4000 5064 4958 3427 3539 1546
5000 5841 5515 3832 3923 1712
The below table 5 demonstrates the various measurement 6000 6618 6072 4237 4307 1878
comparison of the Segmentation recall values between the 7000 7395 6629 4642 4691 2044
existing ISM, DIM G, GM BO, M PSLS and proposed 8000 8172 7186 5047 5075 2210
AISMLA
V. CONCLUS ION
The Image classificat ion is generally seen as a process
implemented on a segmentation basis. But its accuracy can be
overstated if the right choice is not made for the practical
problems that arise in terms of segmentation. So met imes its
accuracy is due to its dot mat rix classificat ion. The current
algorith m based on this has a high degree of accuracy
(96.79%).Thence the proposed Artificial intelligence based

978-1-6654-7971-4/22/$31.00 ©2022 IEEE 1364

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on June 19,2025 at 10:22:08 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Electronics and Sustainable Communication Systems (ICESC 2022)
IEEE Xplore Part Number: CFP22V66-ART; ISBN: 978-1-6654-7971-4

smart machine learn ing algorithm (AI-SM LA) was provide [16] Merkurjev, E., Garcia-Cardona, C., Bertozzi, A.L., Flenner, A., Percus,
A.G.: Diffuse interface methods for multiclass segmentation of high-
better results while compared with the existing an imp roved
dimensional data. Appl. Math. Lett. 33, 29–34 (2014)
segmentation method (ISM), The Diffuse interface models [17] Mohana, J., Yakkala, B., Vimalnath, S., Benson Mansingh, P. M., Srihari,
on graphs (DIM G), Graph M BO method for standoff K., ...&Sundramurthy, V. P. (2022). Application of Internet of Things on
detection (GMBO) and Macroscopic pig mented skin lesion the Healthcare Field Using Convolutional Neural Network Processing.
Journal of Healthcare Engineering, 2022.
segmentation (MPSLS). Thus the separation type can be [18] Merkurjev, E., Kostic, T., Bertozzi, A.L.: An MBO scheme on graphs for
felt to work better. And the process of connecting it is less segmentation and image processing. SIAM J. Imaging Sci. 6(4), 1903–
time consuming and its merg ing method does not take much 1930 (2013)
time. When this is practically possible its speed may vary [19] Dhiman, G., Kumar, A. V., Nirmalan, R., Sujitha, S., Srihari, K.,, ...&
Raja, R. A. (2022). Multi-modal active learning with deep reinforcement
depending on the speed of that hardware processor. learning for target feature extraction in multi-media image processing
applications. Multimedia Tools and Applications, 1-25.
VI. REFERENC E [20] Witten, I.H., Frank, E., Hall, M.A.: Data Mining: Practical Machine
[1] Abbas, Q., Fondon, I., Sarmiento, A., EmreCelebi, M.: An improved Learning T ools and T echniques. Elsevier, Amsterdam (2011)
segmentation method for non-melanoma skin lesions using active
contour model. In: Image Analysis and Recognition, Lecture Notes in
Computer Science, pp. 193–200 (2014)
[2] Bertozzi, A.L., Flenner, A.: Diffuse interface models on graphs for
classification of high dimensional data. Multiscale Model. Simul.
10(3), 1090–1118 (2012)
[3] Merkurjev, E., Sunu, J., Bertozzi, A.L.: Graph MBO method for
standoff detection in hyperspectral video. In: Proc. Int. Conf. Image
Proc, Paris, pp. 689–693 (2014)
[4] Ruuskanen, S., Lehikoinen, E., Nikinmaa, M., Siitari, H., Waser, W.,
Laaksonen, T.: Long-lasting effects of yolk androgens on phenotype
in the pied flycatcher (Ficedulahypoleuca). Behav. Ecol. Sociobiol.
67(3), 361–372 (2013)
[5] vanGennip, Y., Guillen, N., Osting, B., Bertozzi, A.L.: Mean
curvature, threshold dynamics, and phase field theory on finite
graphs. Milan J. Math. 82(1), 3–65 (2014)
[6] Cavalcanti, P.G., Scharcanski, J.: Macroscopic pigmented skin lesion
segmentation and its influence on lesion classification and diagnosis.
In: Color Medical Image Analysis, Lecture Notes in Computational
Vision and Biomechanics, vol. 6. Springer, Berlin (2013)
[7] Chan, T.F., Sanberg, B., Vese, L.A.: Active contours without edges
for vector-valued images. J. Vis. Commun. Image Represent. 11,
130–141 (2000)
[8] Garcia-Cardona, C., Merkurjev, E., Bertozzi, A.L., Flenner, A.,
Percus, A.G.: Multiclass data segmentation using diffuse interface
methods on graphs. IEEE Trans. Pattern Anal. Mach. Intell. 36(8),
1600–1613 (2014)
[9] Kannan, S., Idrees, B. A., Arulprakash, P., Ranganathan, V.,
Udayakumar, E., &Dhinakar, P. (2021). Analysis of convolutional
recurrent neural network classifier for COVID-19 symptoms over
computerised tomography images. International Journal of Computer
Applications in T echnology, 66(3-4), 427-432.
[10] Hu, H., Sunu, J., Bertozzi, A.L.: Multi-class graph Mumford–Shah
model for plume detection using the MBO scheme. In: Tai X.-C., et
al. (eds.)Proceedings of EMMCVPR Hong Kong 2015, Springer
Lecture Notes in Computer Science, vol. 8932, pp. 209–222 (2015)
[11] Gobinathan, B., Mukunthan, M. A., Surendran, S., Somasundaram,
K., Moeed, S. A., Niranjan, P., ...&Sundramurthy, V. P. (2021). A
Novel Method to Solve Real Time Security Issues in Software
Industry Using Advanced Cryptographic T echniques. Scientific
Programming, 2021.
[12] Massone, A.M., Perasso, A., Campi, C., Beltrametti, M.C.: Profile
detection in medical and astronomical images by means of the Hough
transform of special classes of curves. J. Math. Imaging Vis. 51(2),
296–310 (2015)
[13] Arivazhagan, N., Somasundaram, K., Vijendra Babu, D.,
GomathyNayagam, M., Bommi, R. M., Mohammad, G. B.,
...&PrabhuSundramurthy, V. (2022). Cloud-Internet of Health Things
(IOHT) Task Scheduling Using Hybrid Moth Flame Optimization
with Deep Neural Network Algorithm for E Healthcare Systems.
Scientific Programming, 2022.
[14] Merkujev, E., Bae, E., Bertozzi, A., Tai, X.-C.: Global binary
optimization on graphs for classification of high-dimensional data. J.
Math. Imaging Vis. 52(3), 414–435 (2015)
[15] Basker, N., Pradeep, D., Prajapati, R., Arshath Raja, R., ...&Alene, A.
(2022). Healthcare Biclustering-Based Prediction on Gene Expression
Dataset. BioMed Research International, 2022.

978-1-6654-7971-4/22/$31.00 ©2022 IEEE 1365

Authorized licensed use limited to: REVA UNIVERSITY. Downloaded on June 19,2025 at 10:22:08 UTC from IEEE Xplore. Restrictions apply.