IAM Dataset Writer Recognition Using a CNN

Source: Deep Learning on Medium


Handwritten Text Example — is it possible to train an algorithm to identify who wrote it?

During the past few days I have been studying how to do identify the author of a text based on their writing styles. Fortunately, I found the “handwriting_recognition” Jupyter notebook by “Priyanka Dwivedi”, which uses the method explained in the paper “DeepWriter: A Multi-Stream Deep CNN for Text-independent Writer Identification” to solve this problem.

I had some issues while trying to understand the notebook, as it lacks specific instructions on how to setup the dataset and how exactly the data parsing is done. As a result, I decided to rewrite the notebook step-by-step to improve my own understanding of method, and hopefully help others who wish to do the same.

The code and instructions on how to set it up can be found in the “iam_writer_recognition” repository. I have tried to explain each step taken during the process as clearly as possible, as well as to provide visual examples of what the code is doing at each step. Let me know if you have any questions or suggestions on how to improve it.