Research On Progressive Image Generation Technology Based On Generative Adversarial Networks

Image generation algorithm is one of the hot research directions in the field of com-puter vision,and its application is very extensive.At present,the most mainstream image generation algorithms are mainly based on Generative Adversarial Networks and its var-iants.Compared with other image generation algorithms,the image generation algo-rithm based on Generative Adversarial Networks greatly improves the quality of image generation.Image generation technology based on Generative Adversarial Networks has also received extensive attention from the industry.A large number of companies are investing a lot of energy to develop and promote the Generative Adversarial Networks'models,and have achieved certain results.However,image generation technology based on Generative Adversarial Networks still have such problems such as unstable model training process,insufficient image generation diversity,and poor quality of high-reso-lution image generation.This thesis studies the above problems,and the specific work is as follows:1)Aiming at the problem of poor quality of high-resolution image generation and unstable model training process,a Multi-Level Feature Constraints Progressive Gener-ative Adversarial Networks is proposed.The algorithm first extracts the multi-level ab-stract features of the image,and then in the image generation process.Gradually add different levels of abstract feature constraints,and finally generate high-quality high-resolution images under the conditions of multi-layer feature constraints.Compared with the common Generative Adversarial Networks,the Multi-Level Feature Constraints Pro-gressive Generative Adversarial Networks greatly improves the image generation qual-ity,making the generated image more reliable and controllable,especially on high-res-olution images.Ability to produce clearer,higher quality images.2)For the problem of poor quality of high-resolution image generation,a Progres-sive Context-based Constraints Generative Adversarial Networks is proposed,which is different from other progressive generation methods from low resolution to high resolu-tion.The algorithm has a fixed resolution.A small-sized image of the center area is first generated,and then a larger-sized image is continuously generated based on the small-sized image while maintaining the resolution,until a large-sized high-resolution final image is generated.The experimental results on the public dataset show that the algo-rithm can generate large images with high quality and high resolution.