Ancient Yi nationality text character restoration method of a double discriminator GAN

Abstract

The invention provides an ancient Yi nationality text character restoration method of a double discriminator GAN, and the method comprises the steps: constructing a DCGAN network model composed of a Yi nationality text character generator and a Yi nationality text character discriminator, adding a screening discriminator on the basis of the network model, and forming a double discriminator GAN model; through the double discriminator GAN model, acquiring probability distribution of a Yi Yi Yi character image from a multi-source related handwritten Yi Yi Yi, predicting a to-be-restored Yi Yi Yicharacter image through the obtained probability distribution, and completing Yi Yi character restoration according to the predicted image. According to the double discriminator GAN, on the basis of adeep convolutional generative adversarial network, a Yi Yi text screening discriminator is added. Iterative optimization is conducted on a Yi Yi text character generation model through training in two stages and therefore a complete image of a Yi Yi text to be repaired is obtained, and the efficiency of repairing the Yi Yi text characters can be effectively improved.

Description

technical field [0001] The present invention mainly relates to the relevant technical field of Yi language ancient book restoration technology, in particular to a double discriminator GAN ancient Yi character restoration method. Background technique [0002] In China, ancient books in Yi language are increasingly lost and severely damaged. Due to the lack of researchers who are proficient in ancient Yi language, the restoration of ancient books is progressing very slowly. The development of artificial intelligence technology in the field of image text provides the possibility for automatic restoration of ancient books. [0003] In traditional text image restoration, professional researchers use contextual information and perceptual information, that is, through the pixels around the image and the various characteristic elements in the integrated standard text. However, there is no standard text in Yi language for computers to refer to and learn, and there is uncertainty in ...

AI Technical Summary

Problems solved by technology

The generative confrontation network (GAN) proposed by Goodfellow in 2014 has made pioneering progress in the field of image generation. In the process of image generation, the generative confrontation network can better fit the traditional encoder-decoder. data, and the speed is faster, and the generated samples are sharper, but this method also has shortcomings, such as unstable data training, uncontrollable model freedom, and training crashes, etc.

Images