Unsupervised Typography Transfer.

Traditional methods in Chinese typography synthesis view characters as anassembly of radicals and strokes, but they rely on manual definition of the keypoints, which is still time-costing. Some recent work on computer visionproposes a brand new approach: to treat every Chinese character as anindependent and inseparable image, so the pre-processing and post-processing ofeach character can be avoided. Then with a combination of a transfer networkand a discriminating network, one typography can be well transferred toanother. Despite the quite satisfying performance of the model, the trainingprocess requires to be supervised, which means in the training data eachcharacter in the source domain and the target domain needs to be perfectlypaired. Sometimes the pairing is time-costing, and sometimes there is noperfect pairing, such as the pairing between traditional Chinese and simplifiedChinese characters. In this paper, we proposed an unsupervised typographytransfer method which doesn't need pairing.