A Digit Recognition Program

In this project, I used deep neural networks and convolutional neural networks to create a program that prints numbers it observes in real time from images it is given. First, I designed and tested a model architecture that can identify sequences of digits in an image. Next, I trained that model so it could decode sequences of digits from natural images by using the Street View House Numbers (SVHN) dataset. After the model was properly trained, I then tested my model using a program on newly-captured images. Finally, I refined your implementation to also localize where numbers are on the image, and test this localization on newly-captured images.

The main techniques used:

  • Deep Neural Networks
  • Convolutional Neural Networks
  • Adadelta
  • Keras based on TensorFlow

You can see the code(iPython notebook) there.

comments powered by Disqus