KOREAN LETTER HANDWRITING RECOGNITION USING CONVOLUTIONAL NEURAL NETWORK METHOD VGG-16 ARSITEKTUR ARCHITECTURE

Abstract

Handwritten is a unique characteristic because each people has different handwriting. Handwritten can be an object to recognition of someone. In research on handwritten Korean alphabet recognition using the Convolutional Neural Network method with VGG-16 architecture. Data is scanned from 24 Korean handwritten alphabets with 14 kinds of consonants and 10 kinds of vocals on paper with black ink. Data there are two scenarios namely research using original data without binarization and data with binarization which for both scenarios are previously data has been resized. This research uses k-fold cross-validation with a value for k=5 and a confusion matrix. The result showed that both of scenarios are can be recognized with 99,52% accuracy, 95,56% precision, 94,11% recall for first scenario and 99,42% accuracy, 95,94% precision, 93,11% recall for second scenario.