Image Reconstruction from Bernoulli-Dropped Observation Using U-Net

Chunhei Huang

doi:10.54254/2753-8818/2026.HZ28342

Theoretical and Natural ScienceOpen access

Image Reconstruction from Bernoulli-Dropped Observation Using U-Net

Research Article

Open Access

Image Reconstruction from Bernoulli-Dropped Observation Using U-Net

Chunhei Huang ^1*

¹ Vanke Meisha Academic

^*Corresponding author: huangjunxi@stu.vma.edu.cn

Published on 23 October 2025

TNS Vol.145

ISSN (Print): 2753-8826

ISSN (Online): 2753-8818

ISBN (Print): 978-1-80590-467-0

ISBN (Online): 978-1-80590-468-7

Download Cover

Abstract

Image reconstruction under random pixel loss has a significant role in applications such as medical imaging, remote sensing, and lossy transmission. This paper explored the image restoration problem based on the Bernoulli-dropped image, where every pixel has the probability of p to be kept and (1-p) to be removed. This paper modeled the task as a supervised learning problem, utilizing a simple U-Net model (comprising three encoders and decoders) that incorporates skip connections to integrate multi-scale context information and spatial details for image restoration. In this paper, the DIV2K dataset (800 images, grayscale) is applied to the retention rate Random mask with p = 0.3 to generate an observed image that matches its original. The training used the mean square error as the loss function. The result reveals that the model is able to achieve a relatively clear reconstruction effect under the condition of a single input image. It can better preserve edge and texture information, compared to the traditional baseline method. In the end, this paper discusses the issue of choosing and discarding in the network design. Meanwhile, it points out the limitation in extreme pixel loss. At the same time, in the future, potential optimization paths were also mentioned in terms of the improvement of the loss function, the attention mechanism, and the expansion of color images.

Keywords:

Deep learning, U-Net, Image Reconstruction, Bernoulli-dropped Image Reconstruction

View PDF

References

[1]. Quan, W.; Chen, J.; Liu, Y.; Yan, D.-M.; Wonka, P. Deep Learning-based Image and Video Inpainting: A Survey. International Journal of Computer Vision, 2024(accepted). arXiv: 2401.03395; doi: 10.48550/arXiv.2401.03395.O. Elharrouss, N. Almaadeed, S. Al-Maadeed, and R. Akbari, “Deep learning for image inpainting: A survey, ” Pattern Recognition, vol. 122, p. 108341, 2022. doi: 10.1016/j.patcog.2021.108341.

[2]. H. K. Aggarwal, M. P. Mai, and M. Jacob, “A projection-based cascaded U-Net model for magnetic resonance image reconstruction, ” IEEE Transactions on Medical Imaging, vol. 40, no. 5, pp. 1370–1381, 2021. doi: 10.1109/TMI.2020.3047761.

[3]. Aghabiglou, A.; Eksioglu, E. M. Projection-based Cascaded U-Net Model for MR Image Reconstruction. Computer Methods and Programs in Biomedicine, 207: 106151, 2021. doi: 10.1016/j.cmpb.2021.106151.

[4]. Liu, L.; Liu, Y. Load Image Inpainting: An Improved U-Net Based Load Missing Data Recovery Method. Applied Energy, 327: 119988, 2022. doi: 10.1016/j.apenergy.2022.119988.

[5]. Yu, Y.; Zhan, F.; Lu, S.; Pan, J.; Ma, F.; Xie, X.; Miao, C. WaveFill: A Wavelet-based Generation Network for Image Inpainting. In: Proc. IEEE/CVF International Conference on Computer Vision (ICCV), pp. 14114–14123, 2021. doi: 10.1109/ICCV48922.2021.01385.

[6]. T. Xiang, Z. Zhang, R. Wang, C. Zhang, and Y. Wang, “BiO-Net: Learning recurrent bi-directional connections for encoder–decoder architecture, ” Medical Image Analysis, vol. 67, p. 101849, 2020. doi: 10.1016/j.media.2020.101849.

[7]. Xiang, T.; Zhang, C.; Liu, D.; Song, Y.; Huang, H.; Cai, W. BiO-Net: Learning Recurrent Bi-directional Connections for Encoder–Decoder Architecture. In: Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2020, LNCS 12261, pp. 74–84. Springer, 2020. doi: 10.1007/978-3-030-59710-8_8.

[8]. Z. Chen, C. Li, Y. Li, and S. Li, “Self-attention in reconstruction bias U-Net for semantic segmentation of building footprints from high-resolution remote sensing images, ” Remote Sensing, vol. 13, no. 13, p. 2524, 2021. doi: 10.3390/rs13132524.

[9]. R. Timofte, E. Agustsson, L. Van Gool, M. Yang, L. Zhang, B. Lim, et al., “NTIRE 2017 Challenge on Single Image Super-Resolution: Methods and results, ” in Proc. IEEE Conf. Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 1110–1121, 2017.

References

[4]. Liu, L.; Liu, Y. Load Image Inpainting: An Improved U-Net Based Load Missing Data Recovery Method. Applied Energy, 327: 119988, 2022. doi: 10.1016/j.apenergy.2022.119988.

Cite this article

Huang,C. (2025). Image Reconstruction from Bernoulli-Dropped Observation Using U-Net. Theoretical and Natural Science,145,8-16.

Data availability

The datasets used and/or analyzed during the current study will be available from the authors upon reasonable request.

About volume

Volume title: Proceedings of CONF-CIAP 2026 Symposium: International Conference on Atomic Magnetometer and Applications

ISBN: 978-1-80590-467-0(Print) / 978-1-80590-468-7(Online)

Editor: Marwan Omar, Jixi Lu, Mao Ye

Conference website: https://2026.confciap.org/hangzhou.html

Conference date: 30 January 2026

Series: Theoretical and Natural Science

Volume number: Vol.145

ISSN: 2753-8818(Print) / 2753-8826(Online)

© 2024 by the author(s). Licensee EWA Publishing, Oxford, UK. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license. Authors who publish this series agree to the following terms:

1. Authors retain copyright and grant the series right of first publication with the work simultaneously licensed under a Creative Commons Attribution License that allows others to share the work with an acknowledgment of the work's authorship and initial publication in this series.

2. Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the series's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgment of its initial publication in this series.

3. Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See Open access policy for details).