College of Engineering

A Master thesis was discussed in Department of Computer Engineering / College of Engineering at University of Mosul entitled “Handwriting recognition system based on convolutional neural networks” submitted by (Amina Mohammed abd alhameed), Supervised By Assist. Prof. Dr. Mayada Faris Ghanim On Monday, Oct 02, 2023.

In this thesis, deep learning based on a CNN is employed to perform recognition of computer science vocabulary written by participants to build a reliable handwriting recognition system that would aid the educational field.
The model is constructed and validated using previously produced handwritten Computer Science Vocabulary Dataset (HCSVD). The dataset contains the handwritten vocabulary chosen from a dictionary of computer science and engineering and it includes 27682 images for 500 different classes, 120 different students in the College of Engineering participated in the data gathering. After the dataset was compiled, it was prepared to be compatible for use and model training. The preprocessing stage has many phases that contribute to enhancing the images and making them clearer. The proposed CNN model has been evaluated and trained using three different optimizers: Adam, SGD with momentum, and RMSprop.
The results showed that SGD with momentum gave the best results in both accuracy and time consumed, whereby the accuracy was 97% and the recognition time for a single sample was not to exceed 720.63 ms. So this optimizer was later relied upon to study the various hyper parameters like learning rate, batch size, and choice of data size. It was found that the best results were obtained with a learning rate of 0.001, batch size of 32, and data splitting of 70:30. The proposed model also showed good performance in terms of accuracy and time compared to the Alexent model.