Digit Classifier using TensorFlow, Keras and OpenCV Python

By Shrikumar Shankar

This project is an application of a Computer Vision in which Digits from 0 to 9 can be classified and predicted correctly. The project uses Convolutional Neural Networks to predict the images.

This project is an application of Computer Vision in which digits from 0 to 9 can be classified and predicted correctly. The project uses Convolutional Neural Networks to predict the images.

The Model was built using TensorFlow and Keras. TensorFlow is a Machine Learning framework released by Google. It is one of the most popular frameworks used for machine learning. 

The output is shown using OpenCV Python. 

Main Libraries used for Project:-

• Python 3.7
• TensorFlow 2.0.0
• Keras 2.3.1
• OpenCV 4.2.*

Other Libraries include Numpy, scikit-learn, Matplotlib and pickle. 

The Project has 2 files:

1. OCR_CNN_training.py
2. OCR_CNN_test.py

Dataset:

Let us Talk about the Dataset. ORIGINAL DATA SET is from Char74K - http://www.ee.surrey.ac.uk/CVSSP/demos/chars74k/

The dataset consists of images of digits from 0 to 9 in different formats and styles. The images are in RGB format which need to be converted to grayscale later. There are more than 3000 images in the Dataset

OCR_CNN_training.py

This python file is used for training the Deep learning Neural Network. The Images are loaded into the file using OpenCV's imread() function. Each Image is preprocessed and converted to numpy array. The imgaes are normalized for better accuracy. ImageDataGenerator() function is used for Data Augmentation. 

Data augmentation in data analysis are techniques used to increase the amount of data by adding slightly modified copies of already existing data or newly created synthetic data from existing data. It helps reduce overfitting when training a machine learning. It is clesely related to oversampling in data analysis.

The trained model is saved as model_trained_10.p using pickle. The Model has 11 layers and uses Adam Optimizer. The Model trains over 10 epochs with each having 2000 steps. The trained Model is saved using pickle package. It is Later loaded into OCR_CNN_test.py for testing.

OCR_CNN_test.py

This file uses OpenCV and Numpy. The Trained Model is loaded using pickle. Using cv2.VideoCapture() function the video from webcam is captured. The webcam Video feed is converted to Grayscale for better accuracy in prediction. In this the Trained Model is used to predict the Number being displayed on the webcam. The output window shows the Number predicted and the probability of the number displayed being correct.

Note:-

There is a certain threshold set for predicting the number. If the probability is below the threshold the output won't be shown. The Threshold can be modifies by the user.

The Output Directory in the Packet has the sample output for two numbers. The Metrics Directory has the images of Plots, accuracy, accuracy_score achieved etc. 

How to run the project?

1. Create a python virtual environment and Install the dependencies stated above. Make sure the version number is the one mentioned above.
2. Run the OCR_CNN_training.py file for training the model. It might take a lot of time depending on the hardware of your Machine.
3. Once training is completed, Run OCR_CNN_test.py for checking the output. You should Have a webcam or external camera connected to your computer.