Deep Learning Based Hand Writing Word Segmentation

Deep Learning Based Hand Writing Word Segmentation

  • 18 Dec 2016
  • Deep Learning

Word Segmentation

Deep Learning Based Word Segmentation

    • 22 May 2016

The project was based on the paper “A Novel Word Segmentation Method Based on Object Detection and Deep Learning” by Tomas Wilkinson and Anders Brun.[1] It included an implementation of the proposed method for segmenting individual words in a handwritten documents by combining object detection methods along with deep learning trained network. In this project I have focused on the ICDAR2013 [2] dataset for training and testing of the network. The implementation was based on MATLAB,Python and Torch7. MATLAB was used for the preparation of the training data, while the Region Proposal Detector was implemented in Pyton (OpenCV) and the deep network was trained using Torch7 on Google Cloud.

Testing with Other Inputs

Testing with different minimal confidence thresholds, with respect to the network’s output.
These results are based on training over ~20% of the training set only.