Comparisions on KNN, SVM, BP and the CNN for Handwritten Digit Recognition
Wenfei Liu, Jingcheng Wei, Qingmin Meng
Abstract
Handwritten digit recognition technology refers to the automatic identification of handwritten numbers through computers or other equipment, and it has a greater application prospect in letter postal identification and financial statements and bank bill processing. This paper takes the MNIST handwritten digit database as samples, discusses algorithms KNN, SVM, BP neural network, CNN and their application in handwritten digit recognition. In the training process, this work rewrites KNN with Python, SVM with scikit-learn library, and BP, CNN with Tensorflow, and fine-tunes the algorithm parameters to get the best results for each algorithm. Finally, by comparing the recognition rate and recognition duration of the four algorithms, the advantages and disadvantages of the four algorithms in handwriting recognition are analyzed.