REPUBLIQUE DU CAMEROUN REPUBLIC OF CAMEROON

********************* *********************
PAIX-TRAVAIL-PATRIE PEACE-WORK-FATHERLAND
************** **************

SCIENTIFIC PROJECT REPORT

SUBJECT: HANDWRITTEN NUMBER RECOGNITION SYSTEM

GROUP MEMBERS

NAMES IDENTIFIER
DIFFO NZEKOUO HELOISE LAETICIA 21S56721
DONGMO NGUETSOP LAURIE 22S74044
FOUEGANG KEMKA MELI ERIC 22S73742
KONA TETGA ADRIANA ESTHER 22S73865
NDJEUMBA LAURAINE 18S14763
NGNANKOUO MEGNA FADILA 22S76817
NGO MATJABO PRICILLE ANGE ETOILE 22S74294
SONKOUE NGUEYON ELVIRA SOREL 21S57280
TIAKO DJIEUGA LOIC 22S76819
WAFO MURIELLE YOLIE 22S74701
WELADJI NYAMI HILARY FORTUNE 23S91003

Presented on: April, 5 2025

Reviewed by: PhD NGO HANNA

Academic year: 2024/2025

TABLE OF CONTENTS
I. Abstract....................................................................................................................................................................... 5
II. Project’s guide ............................................................................................................................................................ 5
1) Abstract.................................................................................................................................................................. 5
2) Introduction ........................................................................................................................................................... 5
3) Literature Review ................................................................................................................................................... 6
4) Methodology ......................................................................................................................................................... 6
a) Data Collection and Preprocessing .................................................................................................................... 6
b) Model Development ......................................................................................................................................... 7
c) System Architecture .......................................................................................................................................... 9
d) Test and Validation ............................................................................................................................................ 9
5) Results.................................................................................................................................................................. 10
6) Discussion ............................................................................................................................................................ 10
7) Conclusion ........................................................................................................................................................... 10
a) Summary of findings ....................................................................................................................................... 10
b) Future Work .................................................................................................................................................... 10
References ........................................................................................................................................................................ 12
Appendices ....................................................................................................................................................................... 13
1) Confusion Matrix ................................................................................................................................................. 13
2) Accuracy and Loss Graphs ................................................................................................................................... 13
3) System interface Screenshots .............................................................................................................................. 14
 I. List of figures

Figure 1:convolution operation between the image and filter........................................................................................... 6

Figure 2:Padding ................................................................................................................................................................. 7
Figure 3:stride ..................................................................................................................................................................... 7
Figure 4:Layer Max Pooling................................................................................................................................................. 8
Figure 5:Layer Convolutional Network ............................................................................................................................... 8
Figure 6:Layer Flattern ........................................................................................................................................................ 8
Figure 7:Layer Dense .......................................................................................................................................................... 9
Figure 8:Confusion Matrix ................................................................................................................................................ 13
Figure 9:Accuracy and loss Graphs ................................................................................................................................... 14
Figure 10:home page ........................................................................................................................................................ 14
II. Abbreviations
CNN : Convolutional Neural Network

MNIST : Modified National Institute of Standard and Technology

KNN: K-Nearest Neighbors

SVM: Support Vector Machine

 III. Abstract
Handwritten digit recognition plays a crucial role in various industries, such as banking, education, and public
administration, by automating data processing tasks that traditionally require manual input. This project
focuses on developing a web-based application that utilizes machine learning to accurately recognize
handwritten digits. The system is designed to be fast, user-friendly, and efficient, leveraging deep learning
techniques, specifically Convolutional Neural Networks (CNNs), to enhance recognition accuracy. The
objective is to create a reliable tool that can process handwritten digits in real-time, improving efficiency and
reducing human errors in data entry tasks.

IV. Project’s guide

1) Abstract
Handwritten digit recognition is a well-researched problem in the field of artificial intelligence, with
applications in automated data processing and optical character recognition (OCR). This study presents the
development of a machine learning-based web application for digit recognition, utilizing a CNN trained on the
MNIST dataset. The methodology includes data preprocessing, model training, system deployment, and
performance evaluation. The trained model achieves a recognition accuracy of 96.8%. The results demonstrate
the effectiveness of deep learning in improving digit classification accuracy while maintaining computational
efficiency. However, dependence on a stable internet connection remain. Future work will focus on refining
the model, expanding the dataset, and exploring mobile or offline deployment options to enhance usability and
accessibility.

2) Introduction
The increasing reliance on digital automation has highlighted the need for efficient handwritten character
recognition systems, particularly in industries such as banking, education, and public administration, where
handwritten numerical data must be processed quickly and accurately. Traditional manual methods are time-
consuming and prone to errors, while conventional Optical Character Recognition (OCR) systems struggle
with variations in handwriting styles. To address these challenges, this project aims to develop a machine
learning-based web application capable of recognizing handwritten digits with high accuracy. The primary
objectives include designing an intuitive interface for digit input, ensuring a recognition accuracy of over 95%
and validating the system through extensive testing.
3) Literature Review
Several machine-learning approaches have been explored for handwritten digit recognition. Traditional
methods like Support Vector Machines (SVMs) and K-Nearest Neighbors (KNN) have demonstrated moderate
success but struggle with complex handwriting variations. Recent advancements in deep learning, particularly
CNNs, have significantly improved accuracy by automatically learning spatial hierarchies of features. The
MNIST dataset, widely used for benchmarking recognition models, provides a standardized dataset of 70,000
grayscale images of handwritten digits (0-9).

4) Methodology
a) Data Collection and Preprocessing
The dataset used for training and testing the handwritten digit recognition system is the MNIST dataset, which
contains 70,000 grayscale images of digits ranging from 0 to 9. To ensure consistency and compatibility with
the model, all images are resized to 28×28 pixels and converted to grayscale. Preprocessing techniques such
as normalization are applied to scale pixel values between 0 and 1, improving the model’s ability to learn
patterns effectively. Additionally, data augmentation techniques, including rotation and shifting, are used to
enhance model generalization and reduce overfitting.

Figure 1:convolution operation between the image and filter

Figure 2:Padding

Figure 3:stride

b) Model Development
The system is built using a convolutional neural network (CNN), a deep learning architecture well-suited for
image recognition tasks. The CNN consists of multiple convolutional layers that extract spatial features,
pooling layers that reduce dimensionality, and fully connected layers that perform classification. The model is
implemented using TensorFlow and Keras, with the Adam optimizer and categorical cross-entropy as the loss
function to enhance learning efficiency. The training process involves multiple iterations (epochs) with an
adaptive learning rate to optimize performance.
Figure 4:Layer Max Pooling

Figure 5:Layer Convolutional Network

Figure 6:Layer Flattern

Figure 7:Layer Dense

c) System Architecture
The handwritten digit recognition model is deployed as a web-based application to enable user accessibility.
The backend, developed in Python using Flask, processes uploaded images and feeds them into the trained
CNN model for classification. The frontend, designed with HTML, CSS, and JavaScript, provides an intuitive
interface where users can input handwritten digits via image upload or camera capture. The system then
displays the predicted digit along with a confidence score.

d) Test and Validation

The performance of the model is assessed using a combination of evaluation metrics, including accuracy,
precision, recall, F1-score, and a confusion matrix. The model is tested using both the MNIST test dataset and
real-world handwritten digit samples to measure its effectiveness. Additionally, multiple validation techniques,
including unit testing (to verify individual components), integration testing (to assess interactions between
system modules), and user testing (to evaluate overall usability and efficiency), are conducted to ensure the
reliability of the system before deployment.
5) Results
The handwritten digit recognition system demonstrated high accuracy and efficiency in recognizing
handwritten digits. The convolutional neural network (CNN) model achieved an accuracy of 96.8% on the
MNIST test dataset, indicating its ability to effectively classify handwritten digits.
In addition to quantitative performance, the system’s user interface was assessed for usability and
responsiveness. Test users found the web application intuitive and easy to navigate, with smooth image upload
and recognition processes. Overall, the results confirm that the developed system is a reliable and efficient tool
for handwritten digit recognition, with promising applications in banking, education, and administrative
automation.

6) Discussion
The results indicate that the convolutional neural network (CNN) model effectively classifies handwritten
digits with high accuracy, outperforming traditional machine learning approaches such as Support Vector
Machines (SVM) and K-Nearest Neighbors (KNN). The model’s success can be attributed to its ability to
automatically extract spatial hierarchies of features, enabling it to handle diverse handwriting styles.

7) Conclusion

a) Summary of findings
This study successfully developed a machine learning-based handwritten digit recognition system with a web
application interface. By leveraging CNNs, the model achieved a high recognition accuracy of 96.8%,
demonstrating its potential for real-world applications in banking, education, and public administration. The
system met the defined performance criteria and ensuring accuracy in digit classification.

b) Future Work
While the results are promising, future work should focus on further optimizing the model to handle complex
handwriting variations. Additionally, expanding the dataset with diverse handwriting samples and exploring
mobile or offline deployment options could enhance the system’s usability.
References
1. LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document
recognition. Proceedings of the IEEE, 86(11), 2278–2324.
2. Deng, L. (2012). The MNIST database of handwritten digit images for machine learning
research. IEEE Signal Processing Magazine, 29(6), 141–142.
3. Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep Learning. MIT Press.
4. Chollet, F. (2017). Deep Learning with Python. Manning Publications.
5. Geron, A. (2019). Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow (2nd
ed.). O’Reilly Media.
6. MNIST Handwritten Digit Database. (n.d.).
Appendices
1) Confusion Matrix
A confusion matrix showing correct and incorrect predictions across digit classes (0-9), used to analyzed
performance and misclassification trends .

Figure 8:Confusion Matrix

2) Accuracy and Loss Graphs

Graphs showing the model’s accuracy and loss over training epochs, highlighting convergence and model
behaviour.
Figure 9:Accuracy and loss Graphs

3) System interface Screenshots

Images of the web application interface including input/upload screen, prediction output, and user feedback
functionality

Figure 10:home page