SUN’IY INTELLEKT ALGORITMLARI ASOSIDA TASVIRLAR SIFATINI OSHIRISH
Keywords:
sun’iy intellekt, mashinali o‘qitish, Real-ESRGAN, superrezolyutsiya, tasvirni tiklash, kompyuterni ko‘rish, PSNR, SSIM.Abstract
This thesis discusses the problem of improving low-quality
images using machine learning and computer vision technologies. The study
analyzes the Real-ESRGAN algorithm based on generative adversarial networks
and its ability to reduce noise, blur, and compression artifacts. The developed AI
Image Studio platform enables users to upload an image, select an enhancement
mode, compare the output with the original image, and download the improved
result. PSNR and SSIM metrics are considered for evaluating image restoration quality. The proposed approach has practical importance for medical
diagnostics, security systems, digital archives, and satellite image analysis.
References
Bojja, R. R. (2025). A comparative study of supervised and unsupervised learning approaches. International Multidisciplinary Research Journal Reviews.
Mamatov, N. S., & Dusanov, X. T. (2025). Nutq signallari belgilar to‘plamini shakllantirishning MFCC usuli. Современные инновационные исследования: актуальные проблемы и развитие тенденции: решения и перспективы, 179–182.
Mamatov, N. S., Babomuradov, O. J., & Dusanov, X. T. (2024). Automatic speaker recognition through voice using deep neural networks. Journal of Computer Science Engineering and Information Technology Research (JCSEITR), 14(1), 13–20.
Nishonov, A. H. (2023). Neyron tarmoqlari asoslari va ularni amaliyotda qo‘llash. Toshkent: Fan va texnologiya.
Wang, X., et al. (2021). Real-ESRGAN: Training real-world blind super-resolution with pure synthetic data. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops.