Handwritten Multi-Digit Recognition With Machine Learning
Soha Boroojerdi, George Rudolph · 2022 Intermountain Engineering, Technology and Computing (IETC) · 2022
Offline handwritten digit recognition is a well-known problem that remains at best partially solved. This paper presents a study of three different algorithms for offline handwritten multi-digit recognition using the MNIST dataset: Decision Trees, Multilayer Perceptrons and Random Forest. Our results indicate that Random Forest had the best accuracy at 96% with reasonable runtime performance. This kind of study is not novel-however, the authors developed a mechanism for reading multi-digit numbers from image files and webcams that may be of interest.