Machine Learning - Based Handwritten Isolated...

Machine Learning - Based Handwritten Isolated Urdu Character Recognition: Comparative Analysis

Author Name : Sayma Shafeeque A. W. Siddiqui, Rajashri G. Kanke, Ramnath M. Gaikwad, Manasi R. Baheti

ABSTRACT This study describes the implementation and comparison of two commonly used machine learning models for recognizing patterns in a set of isolated handwritten Urdu characters. The first model is built with Support Vector Machines (SVM), whereas the second is built with Convolutional Neural Networks (CNN). The SVM approach did not offer spectacular results; however, the CNN-based implementation achieved an accuracy of 98.82% on the test set, which is the best result produced thus far and serves as a benchmark for the publicly available Urdu dataset HUCD. When comparing the metrics of the results, the resource and time usages of the implementations, as well as the most crucial aspect: accuracy, are taken into account. The procedures were not adjusted to the input in order to generate the most generic outcomes.