README.md

metadata

title: Cheque Dates Predictor LeNet5
emoji: 👁
colorFrom: pink
colorTo: purple
sdk: gradio
sdk_version: 5.44.0
app_file: app.py
pinned: false
license: apache-2.0
short_description: Cheque Dates Predictor using LeNet5 CNN architecture

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

Cheque Dates Predictor using LeNet5

Overview

This project implements a system for extracting and recognizing handwritten dates from cheque images using the LeNet5 Convolutional Neural Network (CNN) architecture. The application automatically locates the date field on a cheque image, segments it into individual digits, and uses a trained CNN model to predict each digit, ultimately reconstructing the complete date in DD/MM/YYYY format.

Features

• Automatic extraction of date fields from cheque images
• Segmentation of date into individual digits (day, month, year)
• Digit recognition using LeNet5 CNN architecture
• Interactive web interface built with Gradio
• Visual feedback showing each step of the process
• Complete date reconstruction in standard format

Model Architecture

The project uses the LeNet5 CNN architecture, which consists of:

• Two convolutional layers with ReLU activation and average pooling
• Three fully connected layers
• Input size of 28x28 grayscale images
• Output of 10 classes (digits 0-9)

The architecture implementation:

class LeNet5(nn.Module):
    def __init__(self):
        super(LeNet5, self).__init__()
        self.conv1 = nn.Conv2d(1, 6, kernel_size=5, stride=1, padding=2)
        self.relu = nn.ReLU()
        self.pool = nn.AvgPool2d(kernel_size=2, stride=2)
        self.conv2 = nn.Conv2d(6, 16, kernel_size=5, stride=1)
        self.fc1 = nn.Linear(16 * 5 * 5, 120)
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)

    def forward(self, x):
        x = self.pool(self.relu(self.conv1(x)))
        x = self.pool(self.relu(self.conv2(x)))
        x = x.view(-1, 16 * 5 * 5)
        x = self.relu(self.fc1(x))
        x = self.relu(self.fc2(x))
        x = self.fc3(x)
        return x

Installation

# Clone the repository
git clone https://github.com/yourusername/Cheque_Dates_Predictor_LeNet5.git
cd Cheque_Dates_Predictor_LeNet5

# Install dependencies
pip install -r requirements.txt

Dependencies

The project requires the following Python packages:

gradio
torch
torchvision
opencv-python
pillow

Usage

Run the application with:

python app.py

The web interface will allow you to:

1. Upload a cheque image
2. View the extracted date region
3. See individual digit predictions
4. Get the complete predicted date

How It Works

The application follows these steps to extract and predict dates from cheque images:

1. Image Preprocessing:

   • The uploaded cheque image is converted to grayscale
   • The image is inverted (255 - pixel value) to enhance digit visibility

2. Date Region Extraction:

   • The system crops the date region using predefined coordinates (x=1790, y=100, width=460, height=80)
   • This region typically contains the handwritten date on standard cheque formats

3. Digit Segmentation:

   • Individual digits are extracted from the date region
   • The system segments the date into 8 parts: 2 for day (D1, D2), 2 for month (M1, M2), and 4 for year (Y1, Y2, Y3, Y4)
   • Each digit is saved as a separate image for processing

4. Digit Recognition:

   • Each digit image is preprocessed (resized to 28x28, normalized)
   • The LeNet5 CNN model predicts the digit value (0-9)

5. Date Reconstruction:

   • Individual digit predictions are combined to form the complete date in DD/MM/YYYY format

Training

The LeNet5 model was trained on the MNIST dataset, which contains 60,000 training images and 10,000 test images of handwritten digits. The training process included:

• Data preprocessing with normalization
• Model training with Adam optimizer and CrossEntropyLoss
• Validation on a separate dataset to prevent overfitting
• Testing on unseen data to evaluate performance

The model achieved high accuracy on digit recognition tasks, making it suitable for cheque date extraction.

Demo

This project is deployed as a Hugging Face Space, providing an interactive demo where users can upload their own cheque images and see the date prediction in action. The demo includes:

• An upload interface for cheque images
• Display of the original and processed images
• Visualization of each extracted digit
• The final predicted date in DD/MM/YYYY format

License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.