Data Science for Business - Group Project

---

---

Setup

1. Make sure you have Python 3.8 or higher installed along with the necessary libraries:

   • Run the following command to install the required libraries :

     pip3 install -r requirements.txt

     OR

     pip install -r requirements.txt

   • OR Run the following commands to install the dependencies:

     !pip install pillow
     !pip install tkinter

---

---

Model Training Guide

Steps to Train the OCR Model

1. Open the OCR Final.ipynb file in Jupyter Notebook or any compatible IDE.
2. Update the paths for:
   • images_folder: The directory containing your labeled images.
   • labels_file: The file mapping image filenames to their respective labels.
3. Run all cells sequentially. The notebook should execute without errors if the dataset is correctly set up.
4. Once training is complete, enjoy your high-performance OCR model!

Pre-Trained Models

If you'd like to skip the training process, you can use one of our pre-trained models:

• retrained_Augmented_LeNet_2.keras: Provides the best balance between accuracy, size, and execution time.
• retrained_Adapted_AlexNet.keras: Achieves the highest accuracy overall.

Simply load these models into your project to start using them immediately.

---

Data Labeling

How to Use the Image Annotation Tool

The following steps will guide you through using the GUI-based Image Annotation Tool for labeling images of digits.

Setup

1. Make sure the setup was done correctly

2. To launch the tool, simply run the provided script.

Instructions for Labeling

Open the labeling_image notebook. Run the code provided in the main cell. Once the tool is running, follow these steps to label your data (also explained in the GUI):

1. Click "Load Images":

   • A file explorer will pop up. Navigate to the folder containing your images and ensure that you enter the folder (not just selecting the folder itself).

2. First Image Appears:

   • You will now see the first image on the screen, ready to be labeled.

3. Draw Orientation Line:

   • Since the digits may have different orientations, you should first draw a line from the center of the leftmost digit to the center of the rightmost digit.
   • Click on the center of the first digit, drag the line, and release at the center of the last digit. A blue line should appear.

4. Confirm Line:

   • A pop-up will ask if the line is correct. If it is, click Yes. If not, click No to redraw.

5. Draw Rectangle Around Each Digit:

   • Now, for each digit, click and drag to draw a rectangle starting from the top-left corner of the digit (reading order). Remember:
     • The top-left corner should always be above and to the left of the bottom-right corner, regardless of the digit's orientation.

6. Label Each Digit:

   • After drawing each rectangle, a pop-up will ask you to enter the digit's label (from 0 to 9). Type the number and press Enter or click yes.

7. Check Label:

   • You should see the labeled information appear in the CSV content area. If you zoom in or out, you’ll see the number appear on the image.

8. Repeat for All Digits:

   • Label all 11 digits in the image. When finished, click Next Image to move to the next one.

Output

Once you're done, the tool will generate:

• A CSV file named annotations.csv containing all labeled data, formatted as follows:
  • "Image Name", "Position", "Label", "Top-Left X", "Top-Left Y", "Bottom-Right X", "Bottom-Right Y"
    where:
    • Image Name corresponds to the original name of the image.
    • Position is the sequential position of the digit in the image.
    • Label is the digit value (0-9).
    • Top-Left X, Y are the coordinates of the top-left corner of the rectangle.
    • Bottom-Right X, Y are the coordinates of the bottom-right corner.
• A "result" folder containing all annotated images with the rectangles and labels drawn.

Navigation Controls

• Zoom In: Use the + key.
• Zoom Out: Use the - key.
• Move Image: Use the arrow keys to pan in any direction.

Quick Summary

1. Load Images → 2) Draw Orientation Line → 3) Draw Rectangle for Each Digit → 4) Enter Digit Label → 5) Move to Next Image

Follow this flow, and you'll quickly annotate all your images!

How to Check the Labeling of your Image

Open the check_labeled_image notebook. Choose the image you want to check and change the name of the image in the corresponding cell. When you changed the name of the image, run the code in the main cell to plot the original image and the labeled image. You can now compare the result of the labeling.

Note: the code in this notebook assume that you have a folder called 'result' with the labeled images created using the instruction above and a folder called 'images' with the original images (before labeling).

How to Analyse Labeling Results

You can find a notebook called 'analysis' which provide analysis about charcater distributions and character positions among the images.