ILLUMINATION INVARIANT FACE RECOGNITION USING DEEP LEARNING FOR ATTENDANCE SYSTEM

Permanasari, Wahyu Melinda (2026) ILLUMINATION INVARIANT FACE RECOGNITION USING DEEP LEARNING FOR ATTENDANCE SYSTEM. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text 22083010043-COVER.pdf Download (3MB)
	Text (BAB I) 22083010043-bab1.pdf Download (270kB)
	Text (BAB II) 22083010043-bab2.pdf Restricted to Repository staff only until 6 March 2028. Download (883kB)
	Text (BAB III) 22083010043-bab3.pdf Restricted to Repository staff only until 6 March 2028. Download (1MB)
	Text (BAB IV) 22083010043-bab4.pdf Restricted to Repository staff only until 6 March 2028. Download (1MB)
	Text (BAB V) 22083010043-bab5.pdf Download (234kB)
	Text (DAFTAR PUSTAKA) 22083010043-daftarpustaka.pdf Download (186kB)
	Text (LAMPIRAN) 22083010043-lampiran.pdf Restricted to Repository staff only until 6 March 2028. Download (301kB)

Abstract

Variations in lighting intensity present a significant challenge in face recognition systems, as they can degrade detection and identification accuracy. This research aims to develop a face recognition system that is robust against lighting fluctuations by integrating Contrast Limited Adaptive Histogram Equalization (CLAHE), MTCNN, FaceNet, and Support Vector Machine (SVM). A primary dataset was collected under controlled conditions from 15 subjects, totaling 675 static images and 135 testing videos, covering variations in low (≤20 lux), moderate (>20–<100 lux), and bright (≥100 lux) lighting, as well as varying distances and facial poses. The evaluation compared three Preprocessing scenarios: original images, CLAHE with a clip limit of 0.001, and CLAHE with a clip limit of 2.0. The results indicate that under low light and at a long distance (100 cm), the system's accuracy without Preprocessing dropped significantly to 77.09% due to MTCNN face detection failures. The implementation of CLAHE significantly improved system performance, achieving 100% voting accuracy and full detection stability (30/30 frames) across all dynamic video testing scenarios. CLAHE with a clip limit of 0.001 was identified as the optimal configuration, yielding an average processing time of 0.8289 seconds per frame, nearly identical to the processing time of original images (0.8288 seconds) while maintaining stable model confidence levels (probability) above 0.94. Furthermore, establishing a probability threshold of 0.74 using the mid-point method proved effective in distinguishing unregistered (unknown) subjects. The best-performing model was integrated into a desktop-based attendance application prototype using PyQt5, featuring real-time liveness detection to verify subject authenticity.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Diyasa, I Gede Susrama Mas	197006192021211009	igsusrama.if@upnjatim.ac.id
Thesis advisor	Pratama, Alfan Rizaldy	199906062024061001	alfan.fasilkom@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Q Science > QA Mathematics > QA76.6 Computer Programming
Q Science > QA Mathematics > QA76.87 Neural computers
T Technology > T Technology (General)

Divisions:

Faculty of Computer Science > Departemen of Data Science

Depositing User:

Wahyu Melinda Permanasari

Date Deposited:

09 Mar 2026 07:29

Last Modified:

09 Mar 2026 07:54

URI:

https://repository.upnjatim.ac.id/id/eprint/50250