Off-line Handwritten Korean Letter using Principle Component Analysis and Back Propagation Neural Network

Abstract

This paper describes a proposed algorithm for recognition of Korean Letters to the Latin language using Principle Component Analysis (PCA) and Back Propagation-Neural Network (BP-NN). The proposed algorithm uses input in the form of image of Korean letters in original 65×65 pixels that is taken from itself. Then, it will be done some processes namely, pre-processing converts image pixel into binary image 15×15 pixels. Further, it transforms from image Red Green Blue (RGB) into binary. Lastly, noise removal from the image. The image will be extracted to produce the image feature. The feature should be processed firstly using Principle Components Analysis (PCA). PCA is used to reduce dimension of image feature before entering classification stage. Classification stage uses a method that called BP-NN. Architecture of ANN uses three hidden layers. Each layer consists of 20, 20 and 5 neurons, and 1 neuron output. The proposed algorithm uses data sampling that is Korean vowels, are obtained from 25 different font types. Next, each font consists of normal sampling and bold sampling. Total data reaches 500 sampling. The data comprises 70% data training and 30% data testing. The result of experiments show that accuracy level is 95%.