Improving the quality of images with focus distortion using convolutional neural networks

Document Type : Research

Authors
Fatemeh Fallahi
Zahra Fallahi
MohammadReza Jafarfard
Babak Zare
10.22034/jmrph.2026.122923.0
Abstract
Defocus distortion is one of the primary factors that degrades the quality of microscopic images, leading to reduced sharpness, loss of fine details, and decreased accuracy in image analysis. Therefore, the development of effective methods for defocus correction is of considerable importance in digital microscopy. In this study, a convolutional autoencoder neural network, consisting of convolutional layers, a compressed bottleneck, and transposed convolutional layers, was designed and trained to correct defocus distortion in digital microscopic images. To this end, a dataset comprising reference images and corresponding defocused images generated using the Fresnel diffraction model was constructed and employed for training and evaluation. The performance of the proposed model was assessed using the Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM). The proposed model achieved an MSE of 0.0008, a PSNR of 30.97 dB, and an SSIM of 0.9839, indicating accurate image reconstruction and excellent structural similarity between the restored and reference images. These results demonstrate that the proposed approach provides an effective solution for defocus correction, improving the quality and sharpness of digital microscopic images.
Keywords
Focus
Fresnel diffraction
deep learning
convolutional neural network
1.        Lee, S., et al., Autofocusing and edge detection schemes in cell volume measurements with quantitative phase microscopy. Optics express, 2009. 17(8): p. 6476-6486.
2.          Kim, T., L. Yong, and Q. Xu. Robust design study on the wide angle lens with free distortion for mobile lens. in AOPC 2017: Optical Storage and Display Technology. 2017. SPIE.
3.          Petersen, B. Optical Anomalies and Lens Corrections Explained. 2016; Available from: https://www.bhphotovideo.com/explora/photography/tips-and-solutions/optical-anomalies-and-lens-corrections-explained.
4.          Islam, J., Towards AI-assisted disease diagnosis: learning deep feature representations for medical image analysis. Ph. D. dissertation, 2019.
5.          Zamir, S. W., et al. Restormer: Efficient Transformer for High-Resolution Image Restoration. CVPR, 5728, 2022.
6.          Liang, J., et al. SwinIR: Image Restoration Using Swin Transformer. ICCV Workshops, 2021.
7.          Mao, X., Shen, C., Yang, Y.-B. Image Restoration Using Very Deep Convolutional Encoder–Decoder Networks with Symmetric Skip Connections. NeurIPS, 29, 2016.
8.          Ronneberger, O., Fischer, P., Brox, T. U-Net: Convolutional Networks for Biomedical Image Segmentation. MICCAI, 234, 2015.
9.          Li, C., et al., Deep learning-based autofocus method enhances image quality in light-sheet fluorescence microscopy. Biomedical Optics Express, 2021. 12(8): p. 5214-5226.
10.       Sun, Y. and G. Mogos, Impact of Visual Distortion on Medical Images. IAENG International Journal of Computer Science, 2022. 49(1).
11.       Sun, Y., Predict the impact of visual distortion on medical images. BSc Thesis, Department of Computer Science and Software Engineering, Xi’an Jiaotong-Liverpool University, Suzhou, China, 2020.
12.       Wang, C., et al., Intelligent autofocus. arXiv preprint arXiv:2002.12389, 2020.
13.       Pedrotti, F.L., L.M. Pedrotti, and L.S. Pedrotti, Introduction to optics. 2017: Cambridge University Press.
14.       Aime, C., E. Aristidi, and Y. Rabbia, The Fresnel diffraction: A story of light and darkness. European Astronomical Society Publications Series, 2013. 59: p. 37-58.
15.       Sara, U., M. Akter, and M.S. Uddin, Image quality assessment through FSIM, SSIM, MSE and PSNR—a comparative study. Journal of Computer and Communications, 2019. 7(3): p. 8-18.
16.       Pinaya, W.H.L., et al., Autoencoders, in Machine learning. 2020, Elsevier. p. 193-208.
17.       Varotto, L., et al., Visual sensor network stimulation model identification via Gaussian mixture model and deep embedded features. Engineering Applications of Artificial Intelligence, 2022. 114: p. 105096.
18.       Zhang, K., et al., Deep image deblurring: A survey. International Journal of Computer Vision, 2022. 130(9): p. 2103-2130.
19.       Zhang, Y. A better autoencoder for image: Convolutional autoencoder. in ICONIP17-DCEC. Available online: http://users. cecs. anu. edu. au/Tom. Gedeon/conf/ABCs2018/paper/ABCs2018
 
Journal Of Modern Research Physics
Volume 11, Issue 1 - Serial Number 20
September 2026
Pages 24-31