Comparación de dos métodos para reconocimiento de dígitos manuscritos fuera de línea

Abstract

In this paper, the results of the comparison between two off-line handwritten digits recognition methods are presented. The first method is a network of perceptrons with which the images were classified after making a comparison by pairs of classes; the second, is a new method that performs a pixel by pixel comparison between the image to be classified, and the reference images. For the tests, a subset of 450 images from the MNIST database was used. Each method was evaluated in two parts: first, with a set of 100 training images, and second, with a set of 350 test images. With the first classifier, an accuracy of 93.86% was obtained, and with the second, an accuracy of 95.14%. After the analysis of the results, it is shown that the second method outperformed the first. The strength of the new method lies mainly in its robustness and execution time.