Chinese character generation is one of the key technologies in Chinese character intelligent design. Most generation methods cannot solve the situation where the generated Chinese character strokes have artifact or ev...
详细信息
ISBN:
(纸本)9789881563903
Chinese character generation is one of the key technologies in Chinese character intelligent design. Most generation methods cannot solve the situation where the generated Chinese character strokes have artifact or even miss. The main reason is that the generator is not strong enough. Therefore, in this paper, we propose a multi-subnet parallel and cascade generator, which combines the conditional generation adversarial network (CGAN) to realize the generation of calligraphy Chinese characters. This method divides a Chinese character generation task into several parts undertaken by each sub-generator, and the output of each sub-generator will contain certain image information, which will increase with the rise of sub-generator hierarchy. It allows each sub-generator to focus on its own task, generating more realistic characters eventually. Finally, the test set performs better than the CGAN with a single generator.
Chinese character generation is one of the key technologies in Chinese character intelligent design. Most generation methods cannot solve the situation where the generated Chinese character strokes have artifact or ev...
详细信息
Chinese character generation is one of the key technologies in Chinese character intelligent design. Most generation methods cannot solve the situation where the generated Chinese character strokes have artifact or even miss. The main reason is that the generator is not strong enough. Therefore, in this paper, we propose a multi-subnet parallel and cascade generator, which combines the conditional generation adversarial network（CGAN） to realize the generation of calligraphy Chinese *** method divides a Chinese character generation task into several parts undertaken by each sub-generator, and the output of each sub-generator will contain certain image information, which will increase with the rise of sub-generator hierarchy. It allows each sub-generator to focus on its own task, generating more realistic characters eventually. Finally, the test set performs better than the CGAN with a single generator.
暂无评论