Open AccessRecognition of MNIST handwritten digits and character set research
Author(s) -
Nataliia Dorosh,
Tatyana Fenenko
Publication year - 2020
Publication title -
ìnformacìjnì tehnologìï v metalurgìï ta mašinobuduvannì
Language(s) - English
Resource type - Journals
ISSN - 2708-0102
DOI - 10.34185/1991-7848.itmm.2020.01.032
Subject(s) - mnist database , pattern recognition (psychology) , digit recognition , artificial intelligence , character recognition , speech recognition , character encoding , mathematics , intelligent word recognition , python (programming language) , handwriting recognition , numerical digit , computer science , character (mathematics) , intelligent character recognition , feature extraction , arithmetic , deep learning , artificial neural network , geometry , image (mathematics) , operating system
The goal of the work is the study of influence of descriptors and reduction of their quantity for recognition of MNIST database of handwritten digits.For recognition of the MNIST digits, a set of 12 descriptors was chosen. Statistical analysis of descriptors was performed. Analysis of descriptors gave the reason to assume, that the fifth, sixth and seventh Hu-moments doesn’t contribute into result of digit recognition. Digit recognition with usage of classifier based on on k-means method with n_neighbors = 10 of Scikit-Learn Python system library was done. Best results using 8 descriptors, excluding the fifth, sixth and seventh Hu-moments and eccentricity. Recognition accuracy was 78.58% compared to 78.14%.