Convolutional Neural Network Untuk Koreksi Deteksi Wajah Berbasis Haar Cascade Dalam Kasus Overdetection Dan Underdetection

Fauzi, Afif Nevandi (2020) Convolutional Neural Network Untuk Koreksi Deteksi Wajah Berbasis Haar Cascade Dalam Kasus Overdetection Dan Underdetection. Undergraduate Thesis thesis, Institut Teknologi Telkom Purwokerto.

Preview	Text ABSTRACT.pdf - Accepted Version Download (6kB) | Preview
Preview	Text ABSTRAK.pdf - Accepted Version Download (7kB) | Preview
	Text COVER.pdf - Accepted Version Download (3MB)
Preview	Text BAB 1.pdf - Accepted Version Download (44kB) | Preview
Preview	Text BAB 2.pdf - Accepted Version Download (356kB) | Preview
Preview	Text BAB 3.pdf - Accepted Version Download (515kB) | Preview
	Text BAB 4.pdf - Accepted Version Restricted to Registered users only Download (506kB)
Preview	Text BAB 5.pdf - Accepted Version Download (7kB) | Preview
Preview	Text DAFTAR PUSTAKA.pdf - Accepted Version Download (126kB) | Preview
	Text LAMPIRAN.pdf - Accepted Version Restricted to Registered users only Download (110kB)

Abstract

One of the research topics in the field of computer vision that is often studied is face detection. Face detection can be done easily by humans, but it is difficult for computers to do because there are several complexities related to location, angle of view, light, and occlusion. The purpose of face detection is to determine the presence or absence of a face in an image, so a method to detect faces is needed. The algorithm that is widely used in face detection is Haar Cascade because it has a high degree of accuracy. However, this algorithm has problems detecting faces. The face detection errors that occur are caused by the similarities in shapes and objects that are similar to faces so that it still needs to be developed. The detection error on the face can be divided into two, namely overdetection and underdetection. Overdetection is a condition where the number of faces detected by the model is more than the actual number of faces. Meanwhile, underdetection is a condition where the number of faces detected by the model is less than the actual number of faces. Based on these problems, this study aims to correct the Haar Cascade algorithm in cases of overdetection and underdetection in face detection to reduce the rate of detection errors. In this study, two scenarios were made, the first scenario was to test face detection using Haar Cascade with 74% accuracy, an average detection speed of 1.11 seconds, an average overdetection case of 18%, and an average underdetection case of 10%. The second scenario is conducting Convolutional Neural Network testing for face detection correction based on Haar Cascade which produces 85% accuracy, an average detection speed of 2.22 seconds, an average overdetection case of 5%, and an average of 16% underdetection cases. Keywords: Haar Cascade; Kesalahan Deteksi; Overdetection; Underdetection; Convolutional Neural Network.

Actions (login required)

View Item