Offline handwritten digit recognition is one of the important tasks in pattern recognition, which is being addressed for several decades. The application of digit recognition lies majorly in areas like postal mail sorting, bank check processing, form data entry, etc. In recent years, research in this area focusses on improving the accuracy and speed of the recognition systems. Many algorithms have been proposed that achieve high recognition rates. In this paper, we propose a highly accurate and fast method using an artificial neural network. Also, we bring out a comparative study of Kirsch directional feature versus the Eigen-based method using two classifiers, Multi-Layer Perceptron (MLP) and Multi-class Support Vector Machine (SVM). Experiments were conducted using the famous MNIST dataset. The proposed method shows a recognition rate of 98.6%.
S N, Muralikrishna
"Eigen-based offline handwritten digit recognition using multi-layer perceptron,"
Manipal Journal of Science and Technology: Vol. 2:
2, Article 7.
Available at: https://impressions.manipal.edu/mjst/vol2/iss2/7