PENGENALAN ANGKA JAWA MENGGUNAKAN EKSTRAKSI CIRI CENTROID FEATURE DAN MULTILAYER PERCEPTRON (MLP)

Abstract

Regional language is a valuable asset of a culture that gave birth to the generation of a nation. In today's digital era has many applications that have been created to recognize a written form of various regional languages. The introduction is done so that the application is able to recognize the writing in the form of an image to be translated into text form.

Wibowo and Wirakusuma in 2013 has conducted a study to identify the letter “ha, na, ca, ra, ka” of aksara java (java script) with the MultiLayer Perceptron method. The introduction of training samples resulted in a success rate of 100%, but the introduction of the test sample resulted in a success rate of 56%. It is caused because the input data are pixels - pixels of the image which has been compressed.

Based on the research conducted experiments to make the first processing on the input data. To recognize the pattern of an image search process performed midpoint (Centroid Feature Extraction) which will then be entered into the MLP. Success rate obtained from centroid feature extraction resulted in a success rate of 100% of the 1000 figure of Java on training samples and 87% of the 500 figure of java on sample testing.