Pengaruh Encoder Residual pada Kinerja U-Net untuk Binarisasi Manuskrip Historis
DOI:
https://doi.org/10.33005/santika.v6i1.957Keywords:
binarisasi citra, manuskrip historis, U-Net, ResNet34, encoder residual, pengolahan citra dokumenAbstract
Binarisasi merupakan tahap penting dalam analisis citra manuskrip historis karena kualitas pemisahan teks dan latar sangat memengaruhi keterbacaan aksara serta proses lanjutan seperti segmentasi dan pengenalan karakter. Penelitian ini menganalisis pengaruh penggunaan encoder residual pada kinerja U-Net untuk binarisasi manuskrip daun lontar dengan membandingkan tiga model, yaitu U-Net dasar, U-Net ResNet34 tanpa pretrained, dan U-Net ResNet34 pretrained. Data yang digunakan berasal dari dua dataset, yaitu AMADI Lontarset dan THPLMD, dengan total 261 citra yang dibagi ke dalam data latih, validasi, dan uji dengan rasio 80:10:10. Seluruh citra diproses menggunakan pendekatan patch-based berukuran 256 × 256 piksel. Evaluasi dilakukan menggunakan metrik IoU, Dice, Recall, Accuracy, SSIM, FSIM, MS-SSIM, GMSD, RMSE, dan PSNR, serta aspek efisiensi komputasi. Hasil penelitian menunjukkan bahwa kedua model U-Net ResNet34 memberikan kinerja lebih baik daripada U-Net dasar pada hampir seluruh metrik. U-Net ResNet34 tanpa pretrained memperoleh hasil terbaik pada IoU sebesar 0,7292, Dice 0,8266, dan Recall 0,8214, sedangkan U-Net ResNet34 pretrained unggul pada Accuracy sebesar 0,9783, SSIM 0,9192, FSIM 0,9385, RMSE 0,1378, dan PSNR 17,7469. Selain itu, kedua varian ResNet34 juga lebih efisien dari sisi waktu pelatihan, inferensi, jumlah parameter, dan penggunaan memori. Temuan ini menunjukkan bahwa encoder residual berperan penting dalam meningkatkan performa binarisasi, sementara pengaruh pretrained bersifat tambahan.
References
I. Yuadi, K. Nisa’, N. U. Nazikhah, Y. A. Halim, A. T. Asyhari, and C.-C. Hu, “A Benchmark Study of Classical and U-Net ResNet34 Methods for Binarization of Balinese Palm Leaf Manuscripts,” Heritage, vol. 8, art. no. 337, 2025, doi: 10.3390/heritage8080337.
N. Otsu, “A Threshold Selection Method from Gray-Level Histograms,” IEEE Transactions on Systems, Man, and Cybernetics, vol. 9, no. 1, pp. 62–66, 1979, doi: 10.1109/TSMC.1979.4310076.
J. J. Sauvola and M. Pietikäinen, “Adaptive document image binarization,” Pattern Recognition, vol. 33, no. 2, pp. 225–236, 2000, doi: 10.1016/S0031-3203(99)00055-2.
O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional Networks for Biomedical Image Segmentation,” in Medical Image Computing and Computer-Assisted Intervention—MICCAI 2015, N. Navab et al., Eds., LNCS, vol. 9351. Cham, Switzerland: Springer, 2015, pp. 234–241, doi: 10.1007/978-3-319-24574-4_28.
P. V. Bezmaternykh, D. A. Ilin, and D. P. Nikolaev, “U-Net-bin: hacking the document image binarization contest,” Computer Optics, vol. 43, no. 5, pp. 825–832, 2019, doi: 10.18287/2412-6179-2019-43-5-825-832.
K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778, doi: 10.1109/CVPR.2016.90.
R. C. Gonzalez and R. E. Woods, Digital Image Processing, 4th ed. Pearson, 2018.
A. N. L. Hanum, A. T. Priyadi, A. N. Hanum, and A. A. Akbar, “Peran library, archives, museums dalam pelestarian naskah kuno di Kalimantan Barat,” Berkala Ilmu Perpustakaan dan Informasi, vol. 18, no. 2, 2022. doi: 10.22146/bip.v18i2.6294.
O. Rizkyantha, R. Rusmiatiningsih, C. Afrina, and M. Oktapia, “Preservasi Naskah Kuno: Strategi Pelestarian Minimal untuk Mempertahankan Warisan Budaya,” Jurnal Pustaka Budaya, vol. 11, no. 1, 2024.
M. Windu et al., “AMADI _ LontarSet : The First Handwritten Balinese Palm Leaf Manuscripts Dataset To cite this version : HAL Id : hal-01389853,” pp. 168–173, 2016, doi: 10.1109/ICFHR.2016.39.
I. Jailingeswari and S. Gopinathan, “Tamil handwritten palm leaf manuscript dataset (THPLMD),” Data Brief, vol. 53, Art. no. 110100, 2024, doi: 10.1016/j.dib.2024.110100.
A. Sulaiman, K. Omar, and M. F. Nasrudin, “Degraded Historical Document Binarization: A Review on Issues, Challenges, Techniques, and Future Directions,” J. Imaging, vol. 5, no. 4, Art. no. 48, 2019, doi: 10.3390/jimaging5040048.
M. Almeida, R. D. Lins, R. Bernardino, D. Jesus, and B. Lima, “A New Binarization Algorithm for Historical Documents,” J. Imaging, vol. 4, no. 2, Art. no. 27, 2018, doi: 10.3390/jimaging4020027.
C. Tensmeyer and T. R. Martinez, “Document Image Binarization with Fully Convolutional Neural Networks,” in Proc. 14th IAPR Int. Conf. Document Analysis and Recognition (ICDAR), 2017, pp. 99–104, doi: 10.1109/ICDAR.2017.25.
J. Calvo-Zaragoza and A.-J. Gallego, “A selectional auto-encoder approach for document image binarization,” Pattern Recognit., vol. 86, pp. 37–47, 2019, doi: 10.1016/j.patcog.2018.08.011.
F. Damayanti, E. M. Yuniarno, and Y. K. Suprapto, “Modified ResUNet Architecture for Binarization in Degraded Javanese Ancient Manuscript,” Math. Model. Eng. Probl., vol. 11, no. 7, pp. 1758–1772, 2024, doi: 10.18280/mmep.110706.
S. Han, S. Ji, and J. Rhee, “Diffusion-Denoising Process with Gated U-Net for High-Quality Document Binarization,” Appl. Sci., vol. 13, no. 20, Art. no. 11141, 2023, doi: 10.3390/app132011141.
L. Zhang, K. Wang, and Y. Wan, “An Efficient Transformer-CNN Network for Document Image Binarization,” Electronics, vol. 13, no. 12, Art. no. 2243, 2024, doi: 10.3390/electronics13122243.
