Structural-parametric Synthesis of Neural Networks for Image Restoration under Synthetic Object-targeted Damage

Authors

DOI:

https://doi.org/10.18372/1990-5548.89.21472

Keywords:

Іtructural-parametric synthesis, image restoration, inpainting, neural networks, genetic algorithm, random erasing, peak signal-to-noise ratio, structural similarity index, encoder-decoder, holdout

Abstract

The article is devoted to the structural-parametric synthesis of neural networks for guided recovery on pairs "damaged - original". The genetic algorithm is used as an optimization algorithm. The training data is synthesized using object-oriented random erasure. The data are built with object-targeted Random Erasing. The train / val / test splits are disjoint: 5362/670/670. On the holdout (n = 670; valid letterbox region, without padding) Wider U-Net yields 25.33 ± 4.50 dB (1.88M; 116.6 GMACs), and Warm Start genetic algorithm yields 25.23 ± 4.39 dB (1.15M; 97.3 GMACs). By the paired Wilcoxon test, the gap behind Wider U-Net is 0.10 dB, and the advantage over Baseline genetic algorithm is 0.25 dB in this run. Both synthesized networks are substantially better than Telea and identity. The level width in the synthesis space is limited to 128 filters, therefore Wider U-Net (256) lies outside it. The result is read as a competitive trade-off, without a claim of absolute PSNR leadership and without generalizing warm start beyond this experiment.

Author Biographies

Olena Chumachenko , National Technical University of Ukraine “Ihor Sikorsky Kyiv Polytechnic Institute”

Doctor of Engineering Science

Professor

Department of Artificial Intelligence

Educational and Research Institute for Applied System Analysis

Anatoliy Kot, National Technical University of Ukraine “Igor Sikorsky Kyiv Polytechnic Institute”

Senior Lecturer

Artificial Intelligence Department

Educational and Research Institute for Applied System Analysis

References

M. Z. Zgurovsky, V. M. Sineglazov, and O. I. Chumachenko, “Formation of Hybrid Artificial Neural Networks Topologies. Artificial Intelligence Systems Based on Hybrid Neural Networks: Theory and Applications,” Studies in Computational Intelligence, vol. 904, pp. 175–232, 2021. Cham: Springer, https://doi.org/10.1007/978-3-030-48453-8_3.

[2] M. Z. Zgurovsky, V. M. Sineglazov, and O. I. Chumachenko, “Classification and Analysis of Multicriteria Optimization Methods. Artificial Intelligence Systems Based on Hybrid Neural Networks: Theory and Applications,” Studies in Computational Intelligence, vol. 904, pp. 59–174, 2021. Cham: Springer. https://doi.org/10.1007/978,-3-030-48453-8_2.

[3] V. Sineglazov, M. Hordun, and S. Junttila, “An intelligent system for determining the degree of tree bark beetle damage based on the use of generative-adversarial neural networks,” Plant-Environment Interactions, vol. 5, no. 6, 2024, Article e70015. https://doi.org/10.1002/pei3.70015.

[4] B. Zoph, V. Vasudevan, J. Shlens, and Q. V. Le, “Learning Transferable Architectures for Scalable Image Recognition,” CVPR, pp. 8697–8710, 2018. https://doi.org/10.1109/CVPR.2018.00907.

[5] E. Real, A. Aggarwal, Y. Huang, and Q. V. Le, “Regularized Evolution for Image Classifier Architecture Search,” AAAI, vol. 33, no. 1, pp. 4780–4789, 2019. https://doi.org/10.1609/aaai.v33i01.33014780.

[6] H. Liu, K. Simonyan, and Y. Yang, “DARTS: Differentiable Architecture Search,” ICLR, 2019. arXiv:1806.09055. https://doi.org/10.48550/arXiv.1806.09055.

[7] H. Pham, M. Guan, B. Zoph, Q. Le, and J. Dean, “Efficient Neural Architecture Search via Parameter Sharing,” ICML, PMLR 80, pp. 4095–4104, 2018. arXiv:1802.03268. https://doi.org/10.48550/arXiv.1802.03268.

[8] L. Xie and A. Yuille, “Genetic CNN,” ICCV, pp. 1388–1397, 2017. https://doi.org/10.1109/ICCV.2017.154.

[9] Z. Lu, I. Whalen, V. Boddeti, Y. Dhebar, K. Deb, E. Goodman, and W. Banzhaf, “NSGA-Net: Neural Architecture Search using Multi-Objective Genetic Algorithm,” GECCO, pp. 419–427, 2019. https://doi.org/10.1145/3321707.3321729.

[10] N. Sinha and K.-W. Chen, “Evolving Neural Architecture Using One Shot Model, GECCO, pp. 910–918. 2021. https://doi.org/10.1145/3449639.3459275.

[11] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A Fast and Elitist Multiobjective Genetic Algorithm: NSGA-II,” IEEE Transactions on Evolutionary Computation, vol. 6, no. 2, 2002, pp. 182–197. https://doi.org/10.1109/4235.996017.

[12] A. Criminisi, P. Perez, and K. Toyama, “Region Filling and Object Removal by Exemplar-Based Image Inpainting,” IEEE Transactions on Image Processing, vol. 13, no. 9, 2004, pp. 1200–1212. https://doi.org/10.1109/TIP.2004.833105.

[13] K. Zhang, W. Zuo, Y. Chen, D. Meng, and L. Zhang, “Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising,” IEEE Transactions on Image Processing, vol. 26, no. 7, 2017, pp. 3142–3155. https://doi.org/10.1109/TIP.2017.2662206.

[14] O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional Networks for Biomedical DOI Image Segmentation,” MICCAI, pp. 234–241, 2015. https://doi.org/10.1007/978-3-319-24574-4_28.

[15] D. Pathak, P. Krahenbuhl, J. Donahue, T. Darrell, and A. A. Efros, “Context Encoders: Feature Learning by Inpainting,” CVPR, pp. 2536–2544, 2016. https://doi.org/10.1109/CVPR.2016.278.

[16] G. Liu, F. A. Reda, J. Shih, T.-C. Wang, A. Tao, and B. Catanzaro, “Image Inpainting for Irregular Holes Using Partial Convolutions,” ECCV, pp. 89–105, 2018. https://doi.org/10.1007/978-3-030-01252-6_6.

[17] A. Lugmayr, M. Danelljan, A. Romero, F. Yu, R. Timofte, and L. Van Gool, “RePaint: Inpainting using Denoising Diffusion Probabilistic Models,” CVPR, pp. 11451–11461. 2022. https://doi.org/10.1109/CVPR52688.2022.01117.

[18] Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image Quality Assessment: From Error Visibility to Structural Similarity,” IEEE Transactions on Image Processing, vol. 13, no. 4, 2004, pp. 600–612. https://doi.org/10.1109/TIP.2003.819861.

[19] T. DeVries and G. W. Taylor, “Improved Regularization of Convolutional Neural Networks with Cutout, arXiv:1708.04552, 2017. https://doi.org/10.48550/arXiv.1708.04552.

[20] Z. Zhong, L. Zheng, G. Kang, S. Li, and Y. Yang, “Random Erasing Data Augmentation,” AAAI, vol. 34, no. 7, pp. 13001–13008, 2020. https://doi.org/10.1609/aaai.v34i07.7000.

[21] R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The Unreasonable Effectiveness of Deep Features as a Perceptual Metric,” CVPR, pp. 586–595, 2018. https://doi.org/10.1109/CVPR.2018.00068.

Downloads

Published

2026-09-18

How to Cite

Chumachenko , O., & Kot, A. (2026). Structural-parametric Synthesis of Neural Networks for Image Restoration under Synthetic Object-targeted Damage. Electronics and Control Systems, 3(89), 45–55. https://doi.org/10.18372/1990-5548.89.21472

Issue

Vol. 3 No. 89 (2026)

Section

COMPUTER SCIENCE

License

Copyright (c) 2026 Олена Іллівна Чумаченко, Анатолій Тарасович Кот

Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 International License.

The scientific journal “Electronics and control systems” adheres to the principles of Open Access and provides free, immediate, and permanent access to all published materials without financial, technical, or legal barriers for readers.

All articles are published in Open Access under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

Copyright

Authors who publish their works in the journal “Electronics and control systems”:

  • retain the copyright to their publications;

  • grant the journal the right of first publication of the article;

  • agree to the distribution of their materials under the CC BY 4.0 license;

  • have the right to reuse, archive, and distribute their works (including in institutional and subject repositories), provided that proper reference is made to the original publication in the journal.