Digital Image Processing for Character Detection of Captcha Login Internet Banking Image using Matching Template
Deni Sutaji, Nadya Husenti · Journal of Physics Conference Series · 2019
Abstract The development of the number of online shopping consumers in Indonesia is increasing. In 2016, research from e-Marketer estimated that it would reach 8.6 million people who shop through the internet. This figure increased from 7.9 million people in the previous year. The increasing number of people who know the internet along with the birth of generation Z (Gen Z) in the digital era makes changes in shopping habits that previously conventionally turned into online. Online transaction is an automatic ticket number that is used to differentiate transactions for each buyer. For example in the purchase transaction of goods in the online shop, each purchase transaction is distinguished through an additional three digits behind the nominal to facilitate payment into the seller’s account. The problem faced by the owner of the online shop is when BRI internet banking login to see account mutations, user authentication by entering the Captcha code. This process is repeated because there is a duration limitation for access to the system. It is necessary to check the account mutation automatically on a computer system that is able to read and recognize the character of the Captcha image. In this study, we propose the Template Matching method in recognizing characters in the captcha image. The dataset contains 300 images taken from the BRI internet banking website. The first step is to convert the extension from .png to .jpg. After the image is converted to extension .jpg the next step is pre-processing to correct the image of the noise that exists. Pre-processing result images that have been fixed are quality, then the separation of objects and background using Otsu. Then the process of labelling and segmentation. The characters are then processed using the Template Matching method so that they can be identified as numbers. The results of the experiments that have been carried out, found that the method we propose is able to recognize characters in the image with an accuracy rate of 90%. Errors in character recognition are due to characters that intersect, so when the results are processed the results do not match the ground truth.