The invention discloses a conditional adversarial network-based copybook binary background noise point removal method. The method comprises the following steps of: 1, establishing a sampling sample; 2, making a manual noise data set according to the sampling sample in the step 1 so as to obtain final training data; 3, training the final training data obtained in the step 2 and updating a discriminator and a generator in sequence; and 4, importing a to-be-processed copybook data sample into the discriminator and the generator in the step 3 to carry out denoising, so as to obtain a noise point-removed copybook. According to the copybook binary background noise point removal method, background noise points caused to literatures by history reasons and improper digital operations can be effectively removed, a denoising operation is provided for digital platform display, and font figures are beautified; and meanwhile, the classification and recognition rate and the cutting precision are improved, and a good using value is provided.