Method and system for generating image by text based on spectrum normalization and comparative learning

The invention discloses a text image generation method and system based on spectrum normalization and comparative learning, and the method comprises the steps: carrying out the comparative learning iteration training of a generator comprising spectrum normalization according to two different key semantic features and word-level text features corresponding to two different text descriptions respectively; utilizing a generator to generate a plurality of images with different resolutions corresponding to the two text descriptions in each iteration training; the discriminator evaluates the similarity between a plurality of images with different resolutions generated by each iteration training and the text description; and iteratively optimizing the generator according to each similarity result until the similarity result meets a condition, obtaining the trained generator at the moment, and generating a target image from the text description by using the trained generator. According to the method, spectrum normali