Machine Learning-based Classification of Indian Caste Certificates using GLCM Features

Computer Science & Engineering: An International Journal (CSEIJ), Vol 15, No 1, February 2025
DOI:10.5121/cseij.2025.15102 9
MACHINE LEARNING-BASED
CLASSIFICATION OF INDIAN CASTE
CERTIFICATES USING GLCM FEATURES
Subramani C 1
, SureshR 2
, Muralidhara B L 1
1
Department of Computer Science and Applications,
Bangalore University, India
2
Department of Statistics, Bangalore University, Bengalore, India
ABSTRACT
In today's world, there is a growing prevalence of fraudulent Caste certificate schemes,
posing a threat to society. It is essential to prevent the issuance of these false documents as
they can potentially disrupt the social order. Accurate classification and verification of
Indian Caste certificates to prevent counterfeiting can ensure equitable access to benefits.
Implementing such digital systems can help thwart these scams, creating a community
devoid of counterfeit documents and promoting a strong social welfare framework. The
detection process includes analyzing scanned copies against authentic references using
image processing methods. This study tackles the issue using texture features extracted
through the Gray Level Co-occurrence Matrix (GLCM) and diverse machine learning (ML)
algorithms. The approach encompasses image pre-processing, GLCM feature extraction,
and the application of classifiers such as K-Nearest Neighbor (KNN), Decision Tree
(DT),Support Vector Machine (SVM),Naive Bayes (NB), andRandom Forest (RF).
Assessments based on accuracy, confusion matrices, and AUC (area under curve) scores
indicate that the Naive Bayes classifier outperforms other methods with 100%accuracy and
a robust AUC score. These findings indicate that integrating GLCM features with ML
algorithms offers a dependable solution for Caste certificate authentication.
KEYWORDS
Image processing, Classification, GrayLevel Co-occurrence Matrix, Machine learning.
1. INTRODUCTION
In India, Caste certificates confirm eligibility for social, educational, and economic benefits,
ensuring access to reservations in education, government employment, and welfare programs.
Nevertheless, extensive forgery undermines these systems, resulting in the unjust distribution of
benefits and placing genuine beneficiaries at a disadvantage.
Authenticating Caste certificates is crucial given their pivotal role. Conventional manual
verification methods are both time-consuming and prone to errors. Image classification presents a
promising solution that automates the verification process. Sophisticated image-processing
methods can discern between genuine and counterfeit certificates, thereby guaranteeing fair
allocation of benefits. This study addresses the task of accurately classifying Indian Caste
certificate images, emphasizing the inefficacy of manual verification. The paper aims to devise an
efficient method utilizing GLCM features and ML algorithms, harnessing texture analysis to
establish a robust verification solution.

10
The field of image classification has attracted considerable attention due to itsinfluential
advancements and successful implementations in domains such as biomedicine, remote sensing,
industry, and many more [1]. In supervised machine learning image classification, two pivotal
processes come into play training and testing. During the training phase, images with known
categories undergo analysis using feature extraction methods to pinpoint significant features,
which are stored as feature vectors. In the testing phase, a new image is introduced to the system
to forecast its class [2]. Feature extraction is vital, as the machine learning algorithm's
performance relies on these features [3].
This research uses theGLCM to derive texture features from Caste certificate images, calculating
energy, correlation, dissimilarity, homogeneity, and contrast across various distances and angles.
These GLCM features will be categorized using ML algorithms. The performance will be assessed
based on accuracy, confusion matrices, and AUC scores. The findings will be scrutinized to
compare classifier performance and highlight their strengths and weaknesses. Since it was first
proposed by Haralick in 1973, the GLCM has been utilized as a texture analysis method to
characterize spatial patterns in images, especially in Gray-level photomicrographs [4]. GLCM has
gained widespread acceptance for image classification through the use of second-order statistical
measures [5]. It serves as a powerful texture descriptor, providing high accuracy and
computational efficiency in comparison to alternative texture extraction techniques [6].
Our proposed work is organized as follows: related work is covered in Section 2, the dataset and
methodology are outlined in Section 3, findings and analysis are discussed in Section 4, and the
study is concluded in Section 5.
2. LITERATURE REVIEW
Research in computer vision has extensively focused on image classification, often entailing the
extraction of handcrafted features and the use of classifiers. Although GLCM features and MLfind
applications in diverse fields, their potential in verifying Caste certificates has not been thoroughly
explored. This study seeks to address this gap by showcasing the effectiveness of GLCM features
in the classification of Caste certificate images.
The statistical texture analysis method known as the GLCM takes into account pixel spatial
relationships. GLCM features offer valuable texture information, including energy, correlation,
dissimilarity, homogeneity, and contrast. Despite its proven efficacy in various classification tasks,
GLCM's utilization in document verification, particularly for Caste certificates, remains limited.
While ML algorithms like Decision Tree, RF,KNN, SVM,and Naive Bayes are well-suited for
image classification, their application to Caste certificate classification using GLCM features lacks
comprehensive documentation.
In research work [7], there is a critical necessity to identify image manipulation, especially
considering the widespread use of images as documentary proof in forensic investigations and
various other fields. The objective of image fraud detection based on pixels is to verify digital
images without requiring prior information about the original image. Images can undergo
tampering through several methods including splicing,resampling, copy-moveand the addition or
removal of objects. Additionally, as mentioned in the work [8], visually detecting image
alterations is extremely challenging for humans. The occurrence of digitally manipulated forgeries
in mainstream media and on the internet is expanding quickly. According to Hany Farid, it was
noted that digital forgeries, despite lacking visual cues, have the potential to modify an image's
underlying statistics. Image forensic tools can be categorized into five primary types: techniques
based on pixels, based on formats, based on cameras, based on physical properties, and techniques
based on geometric features[9].

11
Hayat, Khizar, and Tanzeela Qazi. introduced a forgery detection technique that involves
combiningdiscrete cosine transform (DCT)and discrete wavelet transform (DWT) to reduce
features. In this method, DCT is utilized on blocks generated from the DWT-processed image, and
compared using correlation coefficients. Additionally, as part of the experiments to assess the
detection approach, a mask-based tampering method was formulated and tested. [10]. In their
research, the authors employed an Artificial Neural Network (ANN) classifier to distinguish
tomatoandcotton leaf diseases based on features such as standard deviation, contrast,
homogeneity, energy, mean, correlation, and variance, attaining an overall accuracy of 92.5%
[11]. Another investigation by [12] utilized GLCM texture features in combination with an SVM
classifier to categorize infected tomato leaves, achieving a superior accuracy of 99.83% by
utilizing a linear kernel function. In an investigation conducted by [13], the classification of
sunflower crop diseases, was explored using Multi-Class SVM, NB,KNN,and Multinomial
Logistic Regression (MLR). The classification models employed color and texture features such
asenergy, standard deviation,coarseness, mean, contrast, andhomogeneity. MLR yielded the best
average accuracy of 92.57%, with all classifiers achieving 100% accuracy for healthy leaves.
In their work, A. Singh, Aditi, and Harjeet Kaur. introduced a multi-layer SVM approach with a
linear kernel function to categorize potato leaf diseasesbased on GLCM texture features. The SVM
classifier attained an overall accuracy of 95.99%, along withrecall, precision, and F1-scores of
96.12%, 96.25%, and 96.16%, respectively [14]. In the study conducted by Naveed Iqbal GLCM
features were extracted from grayscale photos collected by a drone. ML techniques such as RF,
NB, SVM, and Neural Network (NN)were employed to classify various crop types. The findings
show that ML algorithms exhibited notably superior performance when utilizing GLCM features,
leading to an overall accuracy enhancement of 13.65% [15].In their work, Mireille Pouyap et al.
suggested the utilization of the GLCM for performing texture analysis on vibration signals within
images. They combined PCA and Sequential Features Extraction (SFE) methods to select
pertinent features. When tested with a multiclass-Naive Bayes classifier, this approach achieved a
success rate of 98.27%, displaying enhanced efficiency and promising outcomes in comparison to
existing methods [16]. Li, Dian, Cheng Wu, and Yiming Wang.suggest an anti-counterfeit method
for iris detection using a binary classification neural network and an enhanced Modified-GLCM.
This method outperforms the leading performance achieved in LivDet-Iris2017 and traditional
texture analysis methods. Furthermore, the study assesses the potential risk of iris adversarial
samples on the iris performance verification system through iris texture extraction [17].
The authors Padmavathi and Maya V. Karkigoal is to derive texture characteristics from brain
tumor cases and categorize them as benign or malignant utilizing,classification phases, feature
extraction, andsegmentation. K-means clustering is employed for segmentation and for selecting
the region of interest. Textural information is collected through GLCM, HOG, and LBP patterns.
The research assesses the accuracy of ANN, k-NN, and SVM classifiers in classifying tumors in
brain MR images. The findings indicate that combining GLCM, LBP, and HOG feature extraction
with an ANN using the ML training algorithm yields higher accuracy and superior performance in
distinguishing benign and malignant tumors compared to other classifiers [18]. In their work,
Barburiceanu, Stefania, Romulus Terebes, and Serban Meza [19] introducea technique that
combines feature vectors from Local Binary Patterns (LBP) and GLCM methods. By employing
classifiers such as SVM, k-NN, and RF, their approach surpasses traditional deep-learning
networks and other customized texture feature extraction methods. Despite a modest number of
images per class, the suggested method enhances discrimination capability and produces
encouraging results. GLCM was utilized which was subsequently condensed to an optimal subset
through PCA. The research revealed that the amalgamation of GLCM with PCA for feature
reduction leads to elevated classification accuracy when employingANN for image categorization
[20].

12
3. PROPOSED WORK
The study exclusively focuses on GLCM features, which capture important texture information.
However, the incorporation of supplementary features or the utilization of alternative machine
learning techniques has the potential to improve classification performance. The development of
image forgery detection systems involves three primary stages: The proposed approach for our
work is illustrated in Figure. 1.
1. Image pre-processing
2. Feature extraction using GLCM statistical features
3. Classification using various ML algorithms.
Figure1:Proposed flowchart for the model
3.1. Data pre-processing
3.1.1. Data Collection
We acquired a thousand images of Karnataka Scheduled Caste and Scheduled Tribe Caste
Certificates from the Atalji Janasnehi Kendra, Nadakacheri Directorate, Karnataka Government.
These images are in jpg format with a resolution of 300 dpi. Abnormal data samples were
generated using Adobe Photoshop to evaluate the proposed ML models. Figure. 2. a and 2. b
displays the samples of both original and fake samples respectively.

13
(a) (b)
3.1.2. Image pre-Processing
The dataset was partitioned with 80% allocated for training and 20% reserved for testing, with
the images resized to 400x400 pixels. Utilizing the original high-resolution images (1653x2339)
in the ML model demands substantial computing resources and can lead to slower training,
potentially resulting in the loss of crucial information. Pre-resizing enhances efficiency and
effectiveness, accelerates computation time, and improves model stability. Rescaling pixel values
between 0 and 1 is necessary for activation function requirements and faster convergence.
Moreover, converting images to grayscale simplifies texture analysis.
3.2. GLCM- Based Feature Extraction
GLCM derives texture features from pre-processed images by determining the frequency of pixel
pairs exhibiting specific values and spatial relationships. GLCM is computed for each image at
various distances (1, 3, 5 pixels) and angles (0, 45, 90, 135 degrees). The extracted features
include contrast, correlation, dissimilarity, energy, and homogeneity, with their equations
discussed below. Energy assesses the uniformity of the texture, while correlation examines the
correlation between a pixel and its neighbor across the complete image. Dissimilarity quantifies
the diversity in graylevel pairs, and homogeneity evaluates the proximity of the concentration of
elements in the GLCM along its diagonal.Contrast gauges the local variations within the GLCM.
Contrast = ∑ ∑ (𝑖 − 𝑗) 𝑃(𝑖, 𝑗) 
Energy = ∑ ∑ 𝑃(𝑖, 𝑗) (2)
Entropy = − ∑ ∑ 𝑃(𝑖, 𝑗) 𝑙𝑜𝑔 𝑃(𝑖, 𝑗) (3)
Dissimilarity = ∑ ∑ |𝑖 − 𝑗| 𝑃(𝑖, 𝑗) (4)
Correlation = ∑ ∑
( ) ( , )
(5)
Where 𝜇 , 𝜇 are the mean and 𝜎 and 𝜎 are the standard deviations of each individual marginal
distribution of 𝑖 and 𝑗.𝑃(𝑖, 𝑗) is the normalized value in the GLCM at position (𝑖, 𝑗) , 𝑁 is the
number of Gray levels in the image.

14
3.3. Classification using Machine Learning Algorithms
The research employs a variety of well-established machine learning algorithms to categorize
GLCM features extracted from images, showcasing their efficacy in diverse classification
endeavors. In this research, five supervised ML classifiers were investigated: RF, SVM, KNN,
DT, and NB. Chosen for their robust performance in image classification, these classifiers
underwent optimization using the grid search method to attain the most favorable outcomes.
The research employs various ML algorithms for classification, with each algorithm being chosen
for its distinctive strengths. RF constructs multiple decision trees and combines them for accurate
and stable predictions. SVM with a linear kernel is utilized for its simplicity and effectiveness,
especially in high-dimensional and binary classification tasks. KNN generates predictions based
on the K in similar instances, making it straightforward to implement and effective for small
datasets. DT models are selected for their ease of interpretation and visualization and their strong
performance with smaller datasets.Lastly, Naive Bayes is a probabilistic model that leverages
Bayes' theorem and assumes that features are independent of one another.
4. RESULTS AND DISCUSSION
In assessing the performance of a trained model, the statistical metrics including recall,precision,
F1-score, accuracy, confusion matrix, and AUROC are employed. Precision evaluates the
percentage of accurate predictionspositives, while recall represents the True Positive Rate (TPR).
4.1. Performance Metrics
The performance of the classifiers is assessed utilizing the following measures:
Accuracy“= 𝑇𝑃 + 𝑇𝑁 (𝑇𝑃 + 𝐹𝑃 + 𝑇𝑁 + 𝐹𝑁)
⁄ “(6)
Precision = 𝑇𝑃 (𝑇𝑃 + 𝐹𝑃)
⁄ (7)
Fallout = 𝐹𝑃 (𝐹𝑃 + 𝑇𝑁)
⁄ (8 )
Recall = 𝑇𝑃 (𝑇𝑃 + 𝐹𝑁)
⁄ (9 )
True Positives are denoted by TP, True Negatives by TN, False Positives by FP, and False
Negatives by FN.
4.2. Accuracy of Proposed Models
Accuracy measures the ratio of correctly identified instances to the overall number of instances
presented in Table1.
Table 1. Comparison of Accuracy of different ML models.
ML algorithm
Classifier
Training
Accuracy
Testing Accuracy
RF 100% 99%
SVM 83% 58%
KNN 95% 81%
DT 100% 98%
NB 98% 100%

4.3. Performance Comparison
The RF and NB classifiers demonstrated
superior performance in the classification
classifier stems from its ensemble
Bayes is a classification method
between features, and is computationally
Decision Trees excel with GLCM
rule-based decisions. In contrast,
encountering challenges when
hyperplane separation.
4.4. Confusion Matrices for Proposed Models
Confusion matrices provide comprehensive insights into classification performance by presenting
the true-positive, false-positive, true
displayed in Table 2. Figure. 3 displays the confusion matrix for
comparing the forecasted labels
data.(a) RF (b) SVM (c) KNN (d) Decision tree (e) Naive Bayes
Table 2. Classificationperfor
Classifier
Performance Comparison
demonstrated the highest accuracy and AUC scores, indicating
classification of Caste certificate images. The robustness
ensemble nature and several decision trees. On the other
method that relies on Bayes' theorem, assuming strong independence
computationally efficient and effective for certain data distributions.
GLCM features by capturing texture patterns and facilitating
contrast, KNN and SVM exhibited relatively lower performance,
dealing with high-dimensional GLCM features and
or Proposed Models
positive, true-negative, and false-negativecounts for each classifier
Figure. 3 displays the confusion matrix for our proposed ML models,
with the real labelsto evaluate the models' performance of test
data.(a) RF (b) SVM (c) KNN (d) Decision tree (e) Naive Bayes
Table 2. Classificationperformance metrics for different ML models.
Classifier TP TN FP FN
RF 70 49 1 0
SVM 70 0 50 0
KNN 70 28 22 0
DT 70 48 2 0
NB 70 50 0 0
(a)
15
indicating their
robustness of the RF
hand, Naive
independence
distributions.
facilitating clear,
performance,
and optimal
negativecounts for each classifier
our proposed ML models,
to evaluate the models' performance of test

(b)
(c)
(d)
16

4.5. Classification Report for Proposed
In machine learning classification,
performance, encompassing metrics
classes. The statistical analysis of
Table 3. Statistical results of the proposed ML models
ML
Algorith
ms
RF
SVM
KNN
DT
NB
4.6. ROC AUC Scores for Proposed Models
The ReceiverOperating Characteristic
classification thresholds. This metric
FPR. Table 4 presents elevated ROC
providing a comprehensive evaluation
curvesof ML models.
Table 4. AUC score of the proposed ML models
Classifier
RF
SVM
KNN
DT
NB
(e)
or Proposed ML Models
classification, a classification report offers an overview of
metrics that evaluate its accuracy in assigning data points
of proposed ML models ispresented in Table 3.
Table 3. Statistical results of the proposed ML models
Precisio
n (%)
Recall
(%)
F1-score
(%)
Accurac
y (%)
99 99 99 99
29 50 37 58
88 78 79 82
99 98 98 98
100 100 100 100
ROC AUC Scores for Proposed Models
Characteristic furnishes a consolidated measure of performance
metric is valuable for assessing the balance between the
ROC AUC scores, indicating enhanced classifier performance
evaluation across all thresholds.Figure.4 displays the
Table 4. AUC score of the proposed ML models
Classifier AUC Score
RF 1.00
SVM 0.98
KNN 0.93
DT 0.98
NB 1.00
17
of a model's
points to specific
performance across all
the TPR and
performance and
displays the AUC-ROC

Figure4:
4.7. Hyperparameter Comparison for GLCM F
In our experiments, the Naïve Bayes classifier delivered flawless accuracy (100%) with a
var_smoothing parameter of 1.0, surpassing all other models. Random Forests
strong performance, achieving 99% accuracy with a max_depth of 10 and n_estimators of 10.
Decision Trees achieved 98% accuracy with a min_samples_split of 5. KNN and SVM were less
effective, with KNN reaching 82% accuracy and SVM 58% accu
parameters. As shown in Table 2.
Table 5. Hyperparameter Comparison for GLCM FEATURES
Algorithm
Random Forests
KNN
Decision Trees
SVM
Naïve Bayes
5. CONCLUSION
This study explores the classification
a variety of ML algorithms. The
the highest accuracies of 99%, 98%,
0.99, 0.98, and 1. These results demonstrate
and counterfeit Caste certificate images.
poorer performance, the GLCM
valuable information for classification.
performance of ensemble methods
study's constraints involve having
Future work should validate the
Figure4:Proposed ROC curve for the model
rparameter Comparison for GLCM Features
var_smoothing parameter of 1.0, surpassing all other models. Random Forests also demonstrated
effective, with KNN reaching 82% accuracy and SVM 58% accuracy using their optimal
Table 5. Hyperparameter Comparison for GLCM FEATURES
Algorithm
Best Parameters
Accuracy
Random Forests max_depth: 10 99
n_estimators: 10
n_neighbors: 3 82
weights: distance
Trees min_samples_split: 5 98
C: 0.1 58
kernel: linear
Naïve Bayes nb__var_smoothing': 1.0 100
'scaler__with_mean': True
classification of Indian Caste certificate images using GLCM
key findings reveal that the RF, DT, and NB classifiers
98%, and 100%, respectively, with corresponding AUC
demonstrate their effectiveness in discriminating between
images. While the KNN and SVM classifiers exhibited
features proved to be effective for texture analysis,
classification. The comparative analysis emphasized the
methods such as RF, DT, and NB. Despite promising
having a relatively small dataset, which may impact generalization.
the findings on larger datasets and explore additional
18
also demonstrated
racy using their optimal
features and
classifiers attained
AUC scores of
between original
exhibited relatively
analysis, providing
the superior
promising results, the
generalization.
additional features or

19
deep learning techniques to further improve classification performance. This researchcontributes
to document verification by using GLCM features and ML algorithms in classifying Caste
certificate images, laying the groundwork for the development of automated systems to identify
forged documents and ensure equitable benefit allocation.
ACKNOWLEDGMENT
The authors extend their appreciation to Directorate Nada Kacheri, Government of Karnataka, for
providing Caste certificate datasets for this research work. The author Subramani C thanks the
University Grants Commission (UGC) New Delhi, India, for providing a fellowship under the
UGC JRF. scheme for research work. (UGC JRF Award letter No.: 210510302177).
REFERENCES
[1] Nath, Siddhartha Sankar, et al. "A survey of image classification methods and techniques." 2014
International conference on control, instrumentation, communication and computational
technologies (ICCICCT). IEEE, 2014.
[2] Thakur, Nupur, and Deepa Maheshwari. "A review of image classification techniques." Int. Res. J.
Eng. Technol 4.11 (2017): 1588-1591.
[3] Humeau-Heurtier, Anne. "Texture feature extraction methods: A survey." IEEE access 7 (2019):
8975-9000.
[4] Jardine, M. A., J. A. Miller, and Megan Becker. "Coupled X-ray computed tomography and grey
level co-occurrence matrices as a method for quantification of mineralogy and texture in
3D." Computers & Geosciences 111 (2018): 105-117.
[5] Pantic, Igor, et al. "Gray level co-occurrence matrix algorithm as pattern recognition biosensor for
oxidopamine-induced changes in lymphocyte chromatin architecture." Journal of theoretical
biology 406 (2016): 124-128.
[6] De Siqueira, Fernando Roberti, William Robson Schwartz, and Helio Pedrini. "Multi-scale gray
level co-occurrence matrices for texture description." Neurocomputing 120 (2013): 336-345.
[7] Ansari, Mohd Dilshad, Satya Prakash Ghrera, and Vipin Tyagi. "Pixel-based image forgery
detection: A review." IETE journal of education 55.1 (2014): 40-46.
[8] Wang, Junwen, et al. "Fast and robust forensics for image region-duplication forgery." Acta
Automatica Sinica 35.12 (2009): 1488-1495.
[9] Farid, Hany. "Image forgery detection." IEEE Signal processing magazine 26.2 (2009): 16-25.
[10] Hayat, Khizar, and Tanzeela Qazi. "Forgery detection in digital images via discrete wavelet and
discrete cosine transforms." Computers & Electrical Engineering 62 (2017): 448-458.
[11] Kumari, Ch Usha, S. Jeevan Prasad, and G. Mounika. "Leaf disease detection: feature extraction
with K-means clustering and classification with ANN." 2019 3rd international conference on
computing methodologies and communication (ICCMC). IEEE, 2019.
[12] Mokhtar, Usama, et al. "SVM-based detection of tomato leaves diseases." Intelligent Systems' 2014:
Proceedings of the 7th IEEE International Conference Intelligent Systems IS’2014, September
24‐26, 2014, Warsaw, Poland, Volume 2: Tools, Architectures, Systems, Applications. Springer
International Publishing, 2015.
[13] Pinto, Loyce Selwyn, et al. "Crop disease classification using texture analysis." 2016 IEEE
International Conference on Recent Trends in Electronics, Information & Communication
Technology (RTEICT). IEEE, 2016.
[14] Singh, Aditi, and Harjeet Kaur. "Potato plant leaves disease detection and classification using
machine learning methodologies." IOP Conference Series: Materials Science and Engineering. Vol.
1022. No. 1. IOP Publishing, 2021.
[15] Iqbal, Naveed, et al. "Gray level co-occurrence matrix (GLCM) texture based crop classification
using low altitude remote sensing platforms." PeerJ Computer Science 7 (2021): e536.
[16] Pouyap, Mireille, et al. "Improved Bearing Fault Diagnosis by Feature Extraction Based on GLCM,
Fusion of Selection Methods, and Multiclass-Naïve Bayes Classification." Journal of Signal and
Information Processing 12.4 (2021): 71-85.

20
[17] Li, Dian, Cheng Wu, and Yiming Wang. "A novel iris texture extraction scheme for iris presentation
attack detection." Journal of Image and Graphics 9.3 (2021): 95-102.
[18] K. Padmavathi and Maya V. Karki. “Texture Feature Extraction and Classification of Brain
Neoplasm in MR Images using Machine Learning Techniques”, International Journal of Recent
Technology and Engineering, Vol. 8, No. 5, pp. 1-9, 2020..
[19] Barburiceanu, Stefania, Romulus Terebes, and Serban Meza. "3D texture feature extraction and
classification using GLCM and LBP-based descriptors." Applied Sciences 11.5 (2021): 2332.
[20] Kumar, Dharmender. "Feature extraction and selection of kidney ultrasound images using GLCM
and PCA." Procedia Computer Science 167 (2020): 1722-1731.

Machine Learning-based Classification of Indian Caste Certificates using GLCM Features

Recommended

More Related Content

Similar to Machine Learning-based Classification of Indian Caste Certificates using GLCM Features (20)

More from CSEIJJournal (20)

Recently uploaded (20)

Machine Learning-based Classification of Indian Caste Certificates using GLCM Features