The U-Net and Pix2Pix GAN Approaches for Image Color Correction in People with Color Blindness

Deke, Milen Juventus Dappa (2026) The U-Net and Pix2Pix GAN Approaches for Image Color Correction in People with Color Blindness. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) 22081010259.-cover.pdf Download (572kB)
	Text (Bab 1) 22081010259.-bab1.pdf Download (161kB)
	Text (Bab 2) 22081010259.-bab2.pdf Restricted to Repository staff only until 15 June 2028. Download (286kB)
	Text (Bab 3) 22081010259.-bab3.pdf Restricted to Repository staff only until 15 June 2028. Download (735kB)
	Text (Bab 4) 22081010259.-bab4.pdf Restricted to Repository staff only until 15 June 2028. Download (1MB)
	Text (Bab 5) 22081010259.-bab5.pdf Download (151kB)
	Text (Daftar Pustaka) 22081010259.-daftarpustaka.pdf Download (208kB)
	Text (Lampiran) 22081010259.-lampiran.pdf Restricted to Repository staff only Download (599kB)

Abstract

Color perception disorders limit an individual’s ability to distinguish colors and accurately recognize objects. This study aims to develop a deep learning-based image color correction system to improve the visual quality of people with color blindness. The research stages include data processing such as data validation, resizing images to 256×256 pixels, converting RGB color space to LMS, and simulating CVD types before splitting the training and test data. The evaluated models include U-Net, Pix2Pix, and a combination of both. Performance was measured using PSNR, SSIM, Delta E metrics, and a user test involving 15 respondents. Experimental results showed that the U-Net model with a learning rate of 0.0003 delivered the best performance with a PSNR of 22.30, SSIM of 0.8405, and Delta E of 10.65. Specifically, the highest model effectiveness was found in the deuteranopia category, while protanopia recorded the lowest performance. Although U-Net was technically superior, user testing showed that accuracy on the original RGB images was still higher than on the corrected images, with a maximum accuracy of 76.51% for deuteranopia. This system has been implemented in a Flask based web application for interactive accessibility. The study concludes that the UNet model is most effective at preserving image quality, but further optimization is needed to improve the consistency of correction results across all user categories.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Muttaqin, Faisal	NIDN0030058602	faisalmuttaqin.if@upnjatim.ac.id
Thesis advisor	Anggraeny, Fetty Tri	NIDN0711028201	fettyanggraeny.if@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA76.87 Neural computers

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Milen Juventus Dappa Deke

Date Deposited:

19 Jun 2026 06:58

Last Modified:

19 Jun 2026 06:58

URI:

https://repository.upnjatim.ac.id/id/eprint/54007