Research On HD Image Transformation Method Based On Deep Interpolation

Face image editing technology which is high-definition face transformation based on computer technology,is an important research area in the direction of computer vision and an important branch of current image editing technology.With the development of computer technology and the popularization of high-definition cameras,face image editing technology is being used for face image editing,face recognition,human-machine verification,human-computer interaction,social security,and face beautification functions.Many urgent needs,the technology for face image editing and face image clarification,are of great significance to the field of face research.As deep learning algorithms are increasingly used in the face area,the ability of convolution neural networks to reduce dimension of faces has made great progress in the transfer of face styles.With the help of deep learning algorithms and deep feature interpolation methods,this paper designs and implements a high-definition human face semantic extraction and semantic editing network for human faces.The main innovations of the article are as follows:1.Designed a new face style migration network,which uses an encoder to decouple face features,uses a style migration module to extract face features,uses a decoder to restore face information,and uses a post-processing module to perform detailed results analysis.Experiments show that the network performs well in the reliability of face style transfer and identity retention.2.The deep learning convolution layer retains the face style parameters.Compared with traditional methods,it can learn more semantic features and make the generated image style details smoother.And due to the characteristics of the convolution neural network,the tested face can be transferred without style alignment.3.Aiming at the problem of loss of semantic information in the resulting part of face style transfer,an end-to-end multitask post-processing module is written to assist the network to recover more details and improve the confidence of the generated pictures.Experiments show that the original image has achieved the basic function of image conversion after deep neural network feature extraction and migration,and this article is more concise,more efficient,uses more network layers,and retains more semantic information,The effect reached a higher quality.