A Study of PyTorch-Based Algorithms for Handwritten Digit Recognition

In this paper, three handwritten digit recognition algorithms based on K-nearest neighbor (KNN), support vector machine (SVM) and convolutional neural network (CNN) are implemented. And compared the performance of the above algorithms in the recognition of handwritten digits using the National Institute of Standards and Technology Modified Database (MNIST) dataset and the USPS dataset. The three algorithms were evaluated in terms of area under the curve (AUC), accuracy (ACC), F1 score (F1), accuracy (PREC) and recall rate (REC), and relative performance index (RPI). We first search for the optimal combination of learning parameters in each algorithm, then evaluate the performance of different algorithms across data sets to obtain the best digit recognition model, and develop a handwritten digit recognition system based on the above algorithm. The system has the functions of numeral input recognition of writing pad, multi-numeral input recognition of picture, and data set making based on input.