Real image reconstruction method based on potential space embedding expansion

The invention relates to a real image reconstruction method based on potential space embedding expansion, which comprises the following steps: firstly, training an intra-domain guide encoder in a supervised manner, mapping a given image into an inherent potential space of a StyleGAN, and introducing pixel loss to maintain visual quality so as to align an encoded latent code in a pixel domain; and introducing perceptual loss d to guide the reversely mapped latent codes to fall into the semantic domain of the StyleGAN space so as to align the coded latent codes in the semantic domain. In order to further improve the quality of the reconstructed image, an embedded optimization algorithm is provided, the expansion of a StyleGAN potential space is realized, the output of an intra-domain guide encoder is used as an initial value for optimization, the latent code generated by the encoder is finely adjusted, and the high-accuracy inverse mapping of a real image is realized. According to the method, the problems that