Research On The Techniques Of Image Emotional Transfer Considering Facial Features

Emotional transfer is designed to change the emotions of the source image.The existing method has some problems.On the one hand,these methods do not consider facial features during the transfer process,which would result in an incorrect emotion assessment for a given image.On the other hand,these methods only transfer the color distribution of the image and does not change the shape features of the image.In order to solve the above problems,this paper studies the image emotion transfer method based on deep learning.Firstly,this paper proposes an emotional transfer method for images with facial features based on convolutional neural network.This method takes into account the facial features.And it establishes a more accurate emotion classification network and analyzes facial emotion characteristics.This paper builds a new emotional database,i.e.,the Face-Emotion database.The database contains the emotional distribution of the images,which is able to match the one-to-many characteristics of images and emotions.Besides,this paper proposes a pre-training emotion network.The network combines local and global features to better learn the details of the image.Finally,the emotional color characteristics of the image are obtained by training data sets.The method not only makes the emotion of the resulting image more consistent with the emotion of the target image,but also ensures the gradient and naturalness of the image.In addition,this paper further explores a new emotional transfer method for images with facial features based on generative adversarial networks.The network consists of two generative adversarial networks that form a mirrored network.Therefore,our network contains two generators and two discriminators.They can share information with each other.This allows the image generated by the generator to be closer to the real image,and the details are better retained.In the loss function,this paper adds facial constraint loss and cycle consistency loss,so that the contour of the face will not be greatly changed.This takes into account the overall contour features of the face as well as the detailed features of the image.Moreover,in order to meet the requirements of the experiment,this paper improves the existing database and divides it into 6 subsets for training.The experimental results show that the new method in this paper is better in terms of detail enhancement and integrity preservation.