Separating Chinese Character from Noisy Background Using GAN

Figure 6

Results of separating printed Chinese characters in Hei font. Columns from left to right: synthesized handwritten/printed overlapped data sample in test set and results of separation of printed characters by CycleGAN, Pix2pix, U-Net, ESRGAN, and DESRGAN. The ground truth and OCR recognition results are also provided. U-Net classifies each pixel into one of two categories and generates a binary image.