To run the program edit the image with the sentence as an arguement to the segmentation method in main.py. A .h5 model must also be given to the test_list_images method then call the method. The final_model4.h5 is built on the EMNIST dataset and has 95% accuracy in recognising capital, small letters and digits on the unseen test set. This is on par with industry standards. The model is built using Keras and TensorFlow as the backend.

This was a project for the Artifical Intelligence and Applied Methods Course at The Royal Institute of Technology KTH.