Development of an Accurate Cheque Reader Based on a Comparative Study of Various Methods of Digit Recognition

Digit recognition is essential in developing machines in the future that can automatically read digits written on cheques. Since the number of applications involving image processing and accurate image classification is increasing, it has become necessary to find a model for a cheque reading system that can read handwritten digits accurately. In this paper, we show how various images of digits can be classified using multiple machine learning models and make a comparison between different models. Five different machine learning algorithms have been used to implement this problem. For each method, the testing handwritten number images are classified, and the variation of the accuracy with the number of principal components considered for our machine learning classifier is reported. It is also possible to figure out the optimal number of principal components for each model. We propose an automated cheque reader based on the comparison of the most accurate model and the optimal principal components. Finally, brief discussions mention using human feedback to improve its performance further.