Implementasi Algoritma Modified Backpropagation untuk Pengenalan Wajah Berbasis Citra

Abstract

In general, the employee attendance system at the office is done by filling out the attendance book using a fingerprint machine where from a security point of view, this system has weaknesses, among others, is that it often experiences human errors such as fingerprint scans that are difficult to accept. This can be caused by abnormal finger surface conditions, such as wet, dirty, too dry, defective finger tips and finally the system refuses, so solving the above problems requires a fast, precise and accurate method. In this study, image-based face recognition was carried out by implementing the Backpropagation algorithm. The facial image dataset used in this study was sourced from the site The Japanese Female Facial Expression (JAFFE) Database with the number of faces to be trained as many as 70 facial images from 10 women with 7 expressions for each face. In this system there are two stages carried out, namely training all face images as a dataset to obtain the weight of each image and the next stage is testing, namely the introduction stage. The experimental results show that the application can read the pixel value of the training image with a maximum error of 0.01, a learning rate of 0.5 and an epoch of 100-1200. The application can perform the introduction with the best results at the 600th epoch with an recognition accuracy value of 85%.