Penggunaan Convolutional Recurrent Neural Network dan RLSA untuk Mengambil Data pada Akta Kelahiran

Abstract

Birth certificate is one of the documents that is mandatory for every citizen to have. This document records the information upon someone’s birth and an official acknowledgement of a country on that person’s existence. Birth certificate is a legal document and is an acceptable form of identification for other documents such as a diploma. As one of Indonesia’s learning institute, Petra Christian University needs its students birth certificate as a solid proof upon their identification and as a base to publish a diploma. The extraction of information is being done manually but with the rapid development of technology, it is now possible to obtain the information within a birth certificate automatically. Research about information extraction on Birth Certificate hasn’t been done yet before, but similar research with the object of Identity Card has been done using Template Matching with the accuracy of 17-39%. This research uses Run Length Smoothing Algorithm and Convolutional Recurrent Neural Network as its primary methods. Run Length Smoothing Algorithm is used to segment words in a birth certificate image. The word in an image will then be translated into a text in string form by Convolutional Recurrent Neural Network. To know which words that contain the wanted information, the sequence of the words and specific keywords are being used. The result of this research will be information upon the full name, birth date, place of birth and the gender of the birth certificate holder. The result from tests that were done is an accuracy of 12.936% upon finding the wanted information and 60.086% for words translation from image to string by CRNN.