In-House Facial Data Collection for Face Recognition Accuracy: Dataset Accuracy Analysis

The face recognition system is a field of research that can help identify a person for either a security system or a search system because faces are distinctive from each person. The development of a human face recognition model requires several main components, such as scale, light, and expression. This study aims to collect accurate samples of Indonesian faces for face detection algorithms. Determining these facial features is challenging to identify Indonesian faces because Indonesia consists of various tribes. In contrast, the classification of tribes in Indonesia consists of Malay and non-Malay tribes. Therefore, this study uses an in-house dataset of Indonesian people's faces which will be tested using the FaceNet algorithm. The results of this study indicate that high accuracy is obtained even when using simple devices. The benefit of this study is that using the in-house method used by researchers can improve the accuracy of test results even though using an in-house dataset.

I. Introduction

Face recognition provides low cost and dependable identification that can be applied in many fields [1]. The human face has unique characteristics that differs others. Therefore facial recognition is a credible source of identification besides fingerprint scanners [2]. Face recognition technology is a biometric technology based on identifying a person's facial features. The facial identification process requires a collection of facial images and automatic recognition equipment to process the images [3]. Several studies discuss the explanation of methods and techniques for developing facial recognition models [4] [5] [6] [7]. However, there is rarely research that includes the creation of a collection of Indonesian facial data, which is the research dataset. One of the studies that used datasets on Indonesian faces was conducted by Wirianto and Tuga Mauritsius, who were labeled Indonesia Labeled Face in the Wild (ILFW).