Hi, I've just finished a basic tutorial on TensorFlow and Keras with MNIST dataset. Now I wonder whether it is possible to apply similar techniques to written letters recognition, or even whole words? Can you advise where to start?
I honestly don't know whether it is the best technique(1), but I'm sure its possible. You can start by just trying with one of these datasets:
Extended MNIST, includes letters: https://www.kaggle.com/crawford/emnist/home
Russian and latin letters: https://www.kaggle.com/gregvial/comnist/home
As for words, I suppose that is a matter of separating a word into letters by image processing. Found these datasets: https://www.quora.com/Computer-Vision-Which-are-some-standard-datasets-for-handwritten-word-image-recognition
If you manage to work it out, you could maybe create a project page and share your knowledge? : )
(1) I don't know much about OCR, I just found https://en.wikipedia.org/wiki/Optical_character_recognition#Techniques Maybe its fun to compare a neural network to k-nearest-neighbour? It seems kNN will involve image processing though...
Just drop us an email to ...
Show more 
edited
Log in with Google