AI-OCR-96

Persian Character based OCR based on Deep Learning. Implemented using Tensorflow.

---

Version 1: Deep Perceptron Neywork

Execute with English MNIST

• The English verion of the code can be easily executed. The Tensorflow library will automatically download the MNIST dataset.
• The entire code is self-contained and is only dependant on a single library, namely Tensorflow.
• The following parameters can be configured inside the code:

learning_rate = 0.004
training_epochs = 30
batch_size = 100

The previous values are the ones proven to be the best and the most efficient based on our experiments.

---

Version 2: Deep Convolutional Neural Network

• Use This line to choose mnist or hoda dataset.
• In order to use Hoda, use this link to download them. Extract the zip archive into the root directory of this project. 4 Files with name test_dump_img_40.npy should be extracted.
• A utility file, parser.cdb.py is provided. You can use it to extract your own cdb files and dump them to .npy.