iitmcvg / ocr-handwritten-text

OCR on NIST Special Datbase 19 using Tensorflow

Handwritten Optical Character Recognition

This is a ongoing project that aims to perform OCR on NIST Special Datbase 19 (handwritten characters) containing 62 classes (uppercase, lowercase and digits) using Tensorflow and Python

Source Programs

• read_and_train.py: Reading and training from the test folder, which contains a small subset of the entire training data
• read3.py: Performs what read_and_train does, with a decaying learning rate
• Boundaries.m and TextSegmentation: Segmenting each letter of the read document into individual letters

Dependencies

• OpenCV Python Bindings
• Numpy
• Tensorflow

ocr-handwritten-text's Issues

what does charsegment.cpp do ?

Database Location

Can you please add a google drive (access based) link for the database on the repo?

Upload it to your SMail account's Google Drive, so that only CVG will have access.

Current Accuracy Status

Why does this model use just a single fully connected layer? And what is the current accuracy?