Inspiration
MasterCard's efforts in promoting digitalized transaction is driven by a pursuit of productivity, sustainability, as well as security. This inspired us to build an automatic card reading program with machine learning techniques. With the model, we aim to promote privacy and efficiency by free human labor from this tedious and sensitive task. Note: We opt in the emerging division.
What it does
The model is able to recognize digits in credit card font at ~100% accuracy in no time. It has the potential to reduce labor cost and it is considered as a black box so that human intervention is not possible.
How we built it
We coded in Python and used the de facto AI package PyTorch, together with NumPy and Pillow. We constructed a fully-connected convolutional neural network to do the task. We collected around 14000 pieces of data and trained the model for 10 epoches.
Challenges we ran into
We had some debate over the choice of loss function and optimizer. We spent some time employing Tensorboard for better visualization of loss convergence.
Accomplishments that we're proud of
Our model's accuracy of recognition reached 100% after around 3 epoches of training.
What we learned
Neural network is a versatile and powerful tool that can free human labor from a variety of tasks.
What's next for Deep Learning for Credit Card Number Recognition
We would realize recognition of the entire sequence of digits on credit cards. To do that, we will develop an AI program that locates the position of card numbers and separates the digits for recognition.
Log in or sign up for Devpost to join the conversation.