Handwritten recognition system based on machine learning

Abstract EN

Background: Handwriting recognition is an important issue nowadays, where handwriting can be a image, document, etc., the ability of a computer to recognize handwritten numbers is very important in more than one application such as translation, reading and number recognition applications.

The proposed project provides a system that recognizes handwritten English numbers, the input data being images downloaded from a global dataset.

The proposed system consists of a number of stages.

The first stage is the preprocessing, which includes resizing of the images to be one size (28 * 28), and then a step (data mapping) is applied.

As for the classification stage, it relied on the use of two algorithms, the KNN algorithm and the neural network (error back propagation).

To start the process of training the selected algorithms, the data was divided into two sets, the training set and the test set.

Two algorithms were used for the purpose of choosing the best of them, by evaluating their performance using a number of evaluation metrics.

Accuracy and Precision were used for the purpose of evaluating the performance of the algorithms.

The performance of the KNN algorithm was 0.94 and 0.942 respectively when k = 4.

While the best performance reached by the neural network mechanism was 0.98673333 and 0.9698, respectively, at epoch=15.

The neural network (error back propagation) is shows the best result in the recognition stage.

Materials and Methods: K-Nearest Neighbors (KNN) technique makes no assumptions about the basic dataset.

It is recognized for its effectiveness and ease of use.

It is a supervised learning algorithm.

To estimate the category of the unlabeled data, a labeled training set containing data points separated into many groups is supplied.

Results: The performance of the KNN model with various values for "K." Since the high value of model accuracy was "0.94", the "4" parameter value is the one that provides the best results and precision was "0.94".

Conclusion: The problem of handwritten recognition needs high accuracy and precision indicators show an accurate description of the performance of the algorithms that were employed in the proposed system.

The two indicators described the performance of the algorithm (KNN), which gave results (0.94 and 0.942).