Research On The Algorithm Of Game-optimized Face Image Editing And Recognition Based On Generative Adversarial Network

With the development of science and technology,artificial intelligence has made great breakthrough and entered a new era in the 21 st century.As a new application direction of artificial intelligence,face attribute transfer has attracted many scholars and Internet companies.Generative Adversarial Network(GAN)applies well in the application of face attribute transfer because it has a good image transfer effect and is in line with human vision.Since it was proposed in 2014,it has received great attention and a large number of experts and scholars improve it all the time.Today it is still a popular research direction in the migration of face attributes.In addition to the migration of face attributes,GANs also have many achievements in text generation,video classification,and image classification.Most GANs map input noise vector to the original picture space through the transposed convolution operation and generate a highly restored picture.The discriminator uses a multi-layer convolutional neural network to determine whether the generated picture comes from the original picture space.However,these methods have their insurmountable problems:(1)Cannot generate pictures according to the example,(2)Cannot process multiple features at the same time,(3)The algorithm runs too long.These problems greatly restrict the use of GAN.So far,these problems are still one of the difficulties and hotspots in the study of GAN.Abouts these problems,an improved adaptive adversarial generation network is proposed,by combining Bayesian equilibrium in dynamic game with incomplete information.By optimizing the generative adversarial networks game process,the different feature generation conditions on the image are identified to form a sub-game between generator and discriminator in the generation process,and the feature coding of different images is observed to judge the source of the image.Secondly,the discriminator does not need to rebuild the neural network to recognize the image features in the model,but calls the encoder in the generator to extract the features twice.The whole game process is actually completed by two parts of the generator.The experiment in this article is based on a certain experimental environment and platform.Through experiments on the Celeb A database,comparing with DAN-GAN algorithm,ELENGANT algorithm and other figure generation algorithms under different evaluation indexes,it is concluded that the Adaptive GAN algorithm can control multiple figures features,and generate figures.At the same time,it shortens the running time of the algorithm and accelerates the convergence speed of the algorithm to a certain extent.