Verification Code Recognition using Machine Learning CNN

Students are required to fill in an additional varification code field before logging in the system since few years ago. Therefore, I am inspired to build up a CNN model for recognizing the varification code.

A sample of the varification code in the course selection system.

Structure

Directories

• training/: Varification code images for training should go here.
• testing/: Other downloaded varification code images for you to test the model.
• model/: The model will be saved in this directory after training.

Python files

• downloadData.py: Download varification code images from the course selection system in my college. It can take a parameter as the number of varification code images to download.
• train.py: Run the training on the CNN model. If the model is already existed in model/, load it and continue training. Otherwise, create a new model.
• predict.py: Predict the varification code from a given varification code images. It can take a parameter as the path to the varification code images.

Usage

Download 100 new varification code images.

python downloadData.py 100

Train the model with images inside training/.

python train.py

Predict a provided varification code image that the model has never seen before.

python predict.py testing/702203.png