Research On The Design Of Visual Illusion Image Generation Based On Creative Adversarial Networks

With the development of artificial intelligence technology,numerous simple and repeated tasks have fully realized the automatic processing by it.Research on making artificial intelligence obtain abstract understanding and creative ability has become the heavy point to march into the next generation of artificial intelligence.As a new medium of art generation,AI art is facing the thorny problems of training data,performance bottleneck and fuzzy evaluation standard boundary when focusing on the cultivation and expression of artificial intelligence creativity.This study,based on the theoretical framework of creative confrontation network(creative adversarial networks,CAN),tries to construct an AI art image generation system centered on human visual perception image cognition from the perspective of generative art.By giving parameterized feedback to artificial intelligence neural network to improve the artistry of creative image,broaden the conceptual extension of AI art and verify the possibility of collaborative artistic creation between human and artificial intelligence.Through literature review and investigation,the definition and participants of AI art are summarized,the types of AI art are divided by analyzing the changes of the relationship between the participants,the logical differences of artificial intelligence algorithms are compared,and the potential of creative confrontation network to enhance the creativity of artificial intelligence is found.Secondly,by combining the requirements of visual recognition and image generation,the target definition of the research system is carried out.The author puts forward the idealized hypothesis for this study by analyzing the characteristics of visual illusion type and the source of ambiguity.Then,by combining creative confrontation network,arousing latent theory and visual error image cognitive model,a visual error image generation system based on creative confrontation network is constructed,and the implementation principle of the system is described from the aspects of evaluation parameters and training logic.After that,a visual-error art image system based on creative confrontation network is created by programming,and the realization of image generation function is verified by the image generated by the system.It shows that the system is effective in the realization of the definition goal.Finally,the overall evaluation of the image generation system proposed in this study is based on the target definition,function construction and performance level.The innovation of this system is to transform the artistic evaluation into creative and ambiguous arousal potential,and to define the internal evaluation criteria of the improved generative adversarial network algorithm by parameterized expression.Although the collective amount of visual error image data is small and the classification quality is low,it has a certain negative effect on the experimental results,but it can still verify the system,especially the cognitive feeling of human as the part of feedback regulation.It plays a positive role in the creative expression of artificial intelligence algorithm.