Chandigarh Engineering College, Jhanjeri, Mohali

(An Autonomous College)

Department of Computer Science and Engineering-Apex

MID-TERM REPORT on

HAND WRITTEN DIGIT RECOGNITION

SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENT FOR THE AWARD OF

THE DEGREE OF

BACHELOR OF TECHNOLOGY

Artificial Intelligence and Data Science

JAN-JUNE, 2025

SUBMITTED BY:

Jayant Goyal Mr. Sandeep Sandhu

Univ. Roll No.: 2130038 Assistant Professor
Semester: 8

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING-Apex

CHANDIGARH ENGINEERING COLLEGE JHANJERI, MOHALI

Affiliated to I.K Gujral Punjab Technical University, Jalandhar

(Batch: 2021-2025)
 Chandigarh Engineering College, Jhanjeri, Mohali
(An Autonomous College)
Department of Computer Science and Engineering-Apex

Table of Contents

S. No. Contents Page No.

1 Introduction 1

2 System Requirement 2-4

3 Progress Till Now 5-6

4 Challenges Faced 7-8

5 References 9
 Chandigarh Engineering College, Jhanjeri, Mohali
(An Autonomous College)
Department of Computer Science and Engineering-Apex

Chapter-1:Introduction

A facial expression is one or more motions or positions of the muscles beneath the skin of the face.
According to one set of controversial theories, these movements convey the emotional state of an
individual to observers. Facial expressions are a form of nonverbal communication. Facial expressions
are vital to social communication between humans. Facial expression classifiers generalizes the learned
features to recognize different expressions from unseen faces. With advancements in computer vision
and deep learning, it is now possible to detect human emotions from images in an improved way.

A Convolutional Neural Network (CNN) is a Deep Learning Algorithm which takes an image as the
input, assigning importance to various objects in the image so that it can differentiate it from others. The
preprocessing required in a CNN model is lower as compared to other classification models. We are
using CNN because it has an ability which automatically detects the important features without any
human supervision.

Emojis are essential to communicate emotion, something that words cannot portray. However, they do
not hold value in the academic world or in a context that demands an objective voice. Emojis are meant
to be fun, light-hearted, and convey a broad range of emotions efficiently and in a way that words
sometimes cannot. Emoji's or avatars are ways to indicate nonverbal cues. These cues have become an
essential part of online chatting, product review, brand emotion, and many more. It also leads to
increasing data science research dedicated to emoji-driven storytelling.

-1-
 Chandigarh Engineering College, Jhanjeri, Mohali
(An Autonomous College)
Department of Computer Science and Engineering-Apex

Chapter-2: System Requirement

2.1 Hardware Requirements
To ensure efficient training, testing, and deployment of the machine learning model for handwritten
digit recognition, the following hardware resources are required:
2.1.1 High-Performance Computing Device:

A laptop or desktop computer equipped with at least an Intel Core i7 or AMD Ryzen 7 processor, 16
GB RAM, and a minimum of 500 GB SSD storage is recommended for local development and testing.
These specifications ensure smooth model training and faster data handling.

2.1.2 GPU for Model Training:

Access to a Graphics Processing Unit (GPU) is crucial for accelerating the training process of
Convolutional Neural Networks (CNNs), which are commonly used for image-based recognition tasks.

Suggested GPUs: NVIDIA GTX 1650 or above, RTX 2060/3060, or Tesla series.

2.1.3 Storage Space:

At least 500 GB of available storage is needed to store image datasets (e.g., MNIST), preprocessed
images, model weights, and log files. SSD is preferred for faster read/write operations.

2.2 Software Requirements

The project uses various software tools and libraries for data handling, model development,
visualization, and deployment.

2.2.1 Operating Systems:

Compatible with:

• Windows 10 or higher
• Linux (Ubuntu preferred for ML environments)
• macOS

Linux is recommended due to its better compatibility and support for open-source machine learning
tools.

-4-
 Chandigarh Engineering College, Jhanjeri, Mohali
(An Autonomous College)
Department of Computer Science and Engineering-Apex

2.2.2 Development Frameworks:

• TensorFlow & Keras: For designing and training Convolutional Neural Networks (CNNs) for
digit classification.
• scikit-learn: Used for preprocessing, splitting datasets, and calculating evaluation metrics like
accuracy or confusion matrix.
• numpy & pandas: Used for data manipulation, numerical computation, and array operations.

2.2.3 Visualization Tools:

• Matplotlib & Seaborn: Used for Exploratory Data Analysis (EDA), visualizing digit samples,
loss and accuracy plots, and performance metrics.

2.3 Dataset Requirements

2.3.1 Data Source:

• The project uses publicly available image datasets such as the FER2013 from Kaggle

2.3.2 Preprocessing Tools:

• OpenCV and Pillow (PIL): These libraries are used for image resizing, grayscale conversion,
normalization, and augmentation. These steps ensure consistent input shape and improve model
generalization by simulating different handwriting styles.

-3-
 Chandigarh Engineering College, Jhanjeri, Mohali
(An Autonomous College)
Department of Computer Science and Engineering-Apex

2.4 Human Resources

2.4.1 Team Members:

The project team should have members with:

• Strong knowledge of Python programming

• Experience in machine learning and deep learning (especially CNNs)
• Familiarity with libraries such as TensorFlow, Keras, and OpenCV
• Basic understanding of model deployment tools (optional for advanced features like web/app
integration)

2.4.2 Mentorship and Guidance:

Guidance and mentorship from faculty advisors or project guides is essential. They provide technical
direction, monitor progress, and ensure that the project meets both academic and performance goals.

-4-
 Chandigarh Engineering College, Jhanjeri, Mohali
(An Autonomous College)
Department of Computer Science and Engineering-Apex

Chapter-3: Progress Till Now

1. Problem Understanding and Objective Definition

• Clearly defined the goal of the project: to develop a system that detects and classifies human
emotions from facial expressions using image or video data.
• Identified the primary emotions to be detected: Happy, Sad, Angry, Surprise, Fear, Neutral, and
Disgust.

2. Literature Review and Research

• Reviewed existing facial emotion detection techniques using machine learning and deep learning
approaches.
• Studied popular models like CNNs (Convolutional Neural Networks), and tools such as
OpenCV, TensorFlow/Keras, and Dlib.

3. Dataset Collection and Preprocessing

• Chosen dataset: [FER-2013 / CK+ / Custom dataset — specify what you used].
• Performed preprocessing tasks such as:
o Grayscale conversion
o Image resizing (e.g., 48x48 or 64x64 pixels)
o Normalization of pixel values
o Label encoding

4. Model Development
• Built an initial deep learning model using Keras/TensorFlow:
o Used multiple convolutional and pooling layers for feature extraction.
o Added dense layers for classification.
o Applied softmax activation in the output layer.
• Implemented model training and validation using training/validation split.

-5-
 Chandigarh Engineering College, Jhanjeri, Mohali
(An Autonomous College)
Department of Computer Science and Engineering-Apex

5. Evaluation Metrics
• Measured performance using accuracy, precision, recall, and confusion matrix.
• Observed overfitting and tuned hyperparameters accordingly.

6. User Interface (Basic)

• Started working on a simple UI using Tkinter / Flask / Streamlit (specify which one).
• Implemented functionality for users to upload an image or access a webcam.

7. Tools and Libraries Used

• Python, OpenCV, TensorFlow/Keras, NumPy, Matplotlib, Pandas, Scikit-learn.

-6-
 Chandigarh Engineering College, Jhanjeri, Mohali
(An Autonomous College)
Department of Computer Science and Engineering-Apex

Chapter-4: Challenges Faced

1. Dataset Quality and Imbalance

• The dataset had an uneven distribution of emotion classes, especially for emotions like
"Disgust" and "Fear", making it difficult to train a balanced model.
• Some images in the dataset were low quality, blurred, or mislabeled, which affected model
performance.

2. Facial Expression Variability

• Emotional expressions vary widely across individuals due to age, ethnicity, and cultural
background, reducing model generalization.
• Subtle emotions like "Neutral" and "Sad" were often misclassified due to minimal differences in
facial features.

3. Overfitting
• The initial model showed high accuracy on the training set but poor generalization on the
validation/test set, indicating overfitting.
• Required experimentation with dropout layers, data augmentation, and regularization techniques
to mitigate it.

4. Real-Time Detection Performance

• While real-time emotion detection using webcam feed was implemented, performance was not
smooth due to:
o Slow frame processing speed
o High CPU/GPU usage
o Difficulty in maintaining detection accuracy across different lighting conditions

-7-
 Chandigarh Engineering College, Jhanjeri, Mohali
(An Autonomous College)
Department of Computer Science and Engineering-Apex

5. Emotion Ambiguity
• Certain facial expressions can represent multiple emotions, leading to ambiguity in
classification (e.g., Surprise vs. Fear).

6. Model Complexity and Training Time

• Deep learning models required substantial training time and computational resources.
• Faced issues running the model efficiently on low-end systems without GPU acceleration.

7. Integration with UI
• Encountered challenges while integrating the backend emotion detection model with a live
interface (Tkinter/Flask/Streamlit), especially handling image inputs and real-time webcam
feed.

-8-
 Chandigarh Engineering College, Jhanjeri, Mohali
(An Autonomous College)
Department of Computer Science and Engineering-Apex

Chapter-5: References

1. FER-2013 Dataset – Kaggle.

https://www.kaggle.com/datasets/msambare/fer2013

2. Goodfellow et al. (2013). Challenges in Representation Learning.

https://arxiv.org/abs/1307.0414

3. Mollahosseini et al. (2017). AffectNet Database.

https://ieeexplore.ieee.org/document/7929250

4. OpenCV Documentation.
https://opencv.org/

5. TensorFlow & Keras Documentation.

https://www.tensorflow.org/
https://keras.io/

6. Python Official Docs.

https://docs.python.org/3/

-9-