Page 24 - 2025S
P. 24
UEC Int’l Mini-Conference No.54 17
nator module. To the best of our research, no [4] Fu, B., Yu, F., Liu, A., Wang, Z., Wen,
previous studies had applied the approach of the J., He, J., Qiao, Y.: Generate like experts:
diffusion approach to the generation of Bengali Multi-stage font generation by incorporat-
fonts. Based on our experimental results, our ing font transfer process into diffusion mod-
proposed method effectively generates Bengali els. In: Proc. of IEEE Computer Vision and
fonts while maintaining their structural sound- Pattern Recognition. pp. 6892–6901 (2024)
ness. The implementation of our proposed dual
aggregation cross-attention has been done suc- [5] Fuad, M.M., Faiyaz, A., Arnob, N.M.K.,
cessfully in terms of combining style information Mridha, M.F., Saha, A.K., Aung, Z.:
on the reference glyph with the content of the Okkhor-diffusion: class guided generation
source glyph. Since our proposed method works of bangla isolated handwritten charac-
on the digitally rendered fonts, we plan to ex- ters using denoising diffusion probabilistic
pand our model in the future to accommodate model (ddpm). IEEE Access (2024)
handwritten glyphs and to improve the perfor-
mance of unseen fonts. [6] He, H., Chen, X., Wang, C., Liu, J., Du,
B., Tao, D., Yu, Q.: Diff-font: Diffusion
model for robust one-shot font generation.
7 Acknowledgment International Journal of Computer Vision
132(11), 5372–5386 (2024)
The Author was supported by a JASSO schol-
arship. The author thanks the Yanai Lab [7] Kong, Y., Luo, C., Ma, W., Zhu, Q.,
for their contributions, computational resources, Zhu, S., Yuan, N., Jin, L.: Look closer
and supportive research environment. Their to supervise better: One-shot font genera-
continuous guidance and encouragement were in- tion via component-based discriminator. In:
strumental in completing this work.This work Proc. of IEEE Computer Vision and Pat-
was supported by JSPS KAKENHI Grant Num- tern Recognition. pp. 13482–13491 (2022)
ber, 22H00548, and JST CRONOS Grant Num- [8] Liu, W., Liu, F., Ding, F., He, Q., Yi, Z.:
ber JPMJCS24K4. Xmp-font: Self-supervised cross-modality
pre-training for few-shot font generation.
In: Proc. of IEEE Computer Vision and
References
Pattern Recognition. pp. 7905–7914 (2022)
[1] Abedin, M.M.h.z., Ghosh, T., Mehrub, T., [9] Lyu, P., Bai, X., Yao, C., Zhu, Z., Huang,
Yousuf, M.A.: Bangla printed character T., Liu, W.: Auto-encoder guided gan for
generation from handwritten character us- chinese calligraphy synthesis. In: 2017 14th
ing gan. In: Soft computing for data ana- IAPR International Conference on Docu-
lytics, classification model, and control, pp. ment Analysis and Recognition (ICDAR).
153–165. Springer (2022) vol. 1, pp. 1095–1100. IEEE (2017)
[2] Azadi, S., Fisher, M., Kim, V.G., Wang, Z., [10] Park, S., Chun, S., Cha, J., Lee, B., Shim,
Shechtman, E., Darrell, T.: Multi-content H.: Few-shot font generation with localized
gan for few-shot font style transfer. In: style representations and factorization. In:
Proc. of IEEE Computer Vision and Pat- Proc. of the AAAI Conference on Artificial
tern Recognition. pp. 7564–7573 (2018) Intelligence. vol. 35, pp. 2393–2402 (2021)
[3] Chang, J., Gu, Y., Zhang, Y., Wang, Y.F., [11] Saha, C., Faisal, R.H., Rahman, M.M.:
Innovation, C.: Chinese handwriting imita- Bangla handwritten basic character recog-
tion with hierarchical generative adversarial nition using deep convolutional neural net-
network. In: Proc. of the British Machine work. In: 2019 Joint 8th International Con-
Vision Conference. p. 290 (2018) ference on Informatics, Electronics & Vision
