Neural Network in C++ With An Object Oriented Approach
SUBMITTED BY: SUBMITTED TO:
Binod Pandey (079bct034) Department of Electronics and Darshan Paudyal (079bct036) Computer Engineering Kapil Upreti (079bct042) ASAR 20 1 1.ACKNOWLEDGEMENT We, the students of the Object-Oriented Programming (OOP) course, want to thank Mr. Daya Sagar Baral for his excellent teaching and constant support during our studies. His deep knowledge of OOP and his engaging teaching style have greatly helped us understand and enjoy this challenging subject.
His ability to explain complicated concepts in a straightforward way has been
incredibly helpful. He created a positive and collaborative classroom environment where we felt comfortable asking questions, sharing ideas, and learning from each other. His patience in answering our questions and his useful feedback have improved our programming skills and boosted our confidence in solving problems. We are very thankful for the time and effort he has put into our education. His guidance has had a lasting impact on us, shaping how we approach programming and our future goals. Thank you, Sir, for being an outstanding teacher and mentor. We will always appreciate the knowledge and wisdom you have shared with us.
Our sincere thanks also go to our peers and the Department of Computer Engineering for providing us with the resources and environment to pursue this project.
Sincerely,
Binod Pandey
Darshan Paudyal
Kapil Upreti
2 2.TABLE OF CONTENT 1.ACKNOWLEDGEMENT .................................................................................... 2
2.TABLE OF CONTENT ........................................................................................ 3
3 3.INTRODUCTION Recognizing handwritten characters is important for many tasks today. These tasks include turning handwritten documents into digital files, automatically processing forms, and translating languages. While there are many tools for recognizing English handwriting, there is a big gap when it comes to recognizing Nepali characters. This makes it hard to digitize important documents and automate form processing in Nepali.
This project aims to fill this gap by creating a system to recognize Nepali characters, both letters and numbers. We will use an artificial neural network (ANN), which is a type of computer system that learns to recognize patterns, much like the human brain. We will build this ANN from scratch using the C++ programming language.
By developing this system, we hope to make it easier to digitize Nepali texts,
process forms automatically, and improve language translation services. This will help preserve and promote the Nepali language in the digital world.
4 4.OBJECTIVES ▪ Gain a solid understanding of Object-Oriented Programming (OOP) concepts and apply them practically in a project. ▪ Explore and master the foundational aspects of the C++ programming language. ▪ Learn how to maximize code reuse by creating and using custom header files. ▪ Get familiarity with the SFML (Simple and Fast Multimedia Library) for developing software that interacts with multimedia components on computers. ▪ Develop programs that are not only effective in achieving their goals but also efficient in terms of improving both time and space resources.
5.EXISTING SYSTEM Currently, there are limited systems available for Nepali character recognition. Most existing OCR systems focus on English characters, and those that do handle Nepali characters are often proprietary and not open source. Furthermore, these systems typically rely on pre-built machine learning libraries rather than implementing the neural network from scratch.
5 6.PROPOSED SYSTEM 6.1 DESCRIPTION The proposed system aims to develop an Artificial Neural Network (ANN) from scratch using C++ with an object-oriented approach. This ANN will be specifically tailored for the recognition of Nepali characters, including both letters and numbers. The system will encompass several critical components: a dataset of Nepali characters, a training module to train the ANN, a testing module to evaluate the ANN’s performance, and a user interface for real-time recognition of handwritten Nepali characters.
System Components
1. Dataset of Nepali Characters:
o Collection: Images of Nepali characters, sourced from Kaggle. o Preprocessing: Binarization, normalization, and noise reduction. o Storage: Organized for efficient access during training and testing.
2. Training Module: o Architecture Design: Multiple layers optimized for character recognition. o Algorithm Implementation: Backpropagation and gradient descent from scratch. o Training Process: Loading dataset, feeding images, calculating errors, adjusting weights. o Parameter Tuning: Optimizing learning rate, batch size, and epochs. o Validation: Using a validation set to monitor training and prevent overfitting.
3. Testing Module: o Performance Metrics: Accuracy, precision, recall, and F1 score. o Confusion Matrix: Visualizing performance across different characters.
4. User Interface for Real-Time Recognition:
o Input Mechanism: Drawing pad or image upload. o Real-Time Processing: Immediate character recognition. o Output Display: Showing recognized characters with confidence scores.
6 6.2 SYSTEM BLOCK DIAGRAM
7 7. METHODOLOGY 7.1 Dataset Creation: ▪ Collect images of Nepali characters. ▪ Preprocess images (grayscale conversion, normalization, etc.).
7.2 Image Processing:
▪ Apply techniques such as binarization and noise reduction. ▪ Use OpenCV for dealing with Images in C++.
7.3 Feature Extraction:
▪ Extract features from preprocessed images to be used as input for the ANN.
7.4 ANN Development:
▪ Implement the neural network architecture (input layer, hidden layers, output layer) from scratch in C++. ▪ Train the ANN using the dataset. ▪ Optimize the ANN through techniques like backpropagation and gradient descent.
7.5 Testing and Evaluation:
▪ Test the trained ANN on a separate dataset. ▪ Evaluate performance using metrics such as accuracy, precision, and recall.
7.6 User Interface Development:
▪ Create an interface for users to input and recognize Nepali characters in real-time.
8 8. PROJECT SCOPE The project scope includes the development of an ANN for Nepali character recognition with a primary focus on implementing the neural network from scratch. This project will serve as a foundational tool for our learning, research and development in the field of Nepali OCR. Potential extensions could include expanding the character set, improving recognition accuracy, and deploying the system as a web or mobile application.
9.PROJECT SCHEDULE
TASK DURATION START END DATE
DATE Data collection 1 week 03 - 09 03 - 15
System Design ½ week 03 - 16 03 - 18
ANN Implementation ½ week 03 - 19 03 - 22
GUI Development 1 week 03 – 23 03 - 28
Training and testing ½ week 03 - 29 04 - 01
Iteration and Optimization 1 week 04 - 02 04 - 07
Final Documentation ½ week 04 - 08 04 - 11
Project Submission - - 04 - 13
This proposal outlines our plan to develop a Nepali Characters Recognition system using Artificial Neural Networks. By adhering to this structured approach, we aim to achieve a functional and efficient solution for Nepali script recognition.
