Main Article Content
The recognition of cursive or a running hand script is considered a difficult task in character recognition because it has a different representation style. Urdu originated from Arabic script, which is why it is much closer to Arabic script, has similar challenges and complications but with more intensity level. There are different styles of writing Urdu, but commonly Urdu alphabet is written in Nastalik script. This research work done on Convolutional Neural Networks with Mobile Net architecture with a Machine learning technique, i.e. Transfer Learning, makes it much easier. It is a technique where the model is developed for one task and then re-used as a starting point for different tasks. There are 603 images of Urdu Handwritten Words with 44 classes written by a different writer in datasets. The size of all images is 64*64, which is trained through the transfer learning technique. Code written in python using different python libraries like Keras, Ski learn, Numpy, etc., and accuracy is 90%, which is very efficient, later discussed in the last sections.
Copyright (c) 2021 Technical Journal
The author transfers all copyright ownership of the manuscript entitled (title of article) to the Technical Journal in the event the work is published.