Research on Image Style Convolution Neural Network Migration Based on Deep Hybrid Generation Model

Abstract

The main content of image style transfer transforms the image style from one region to another. This task puts forward new needs for the traditional convolutional neural network architecture. Therefore, a deep hybrid generation model is usually used in the study of processing image style transfer. Image style transfer aims to transform the image into a new idea by image generation. This paper proposes an image-style convolution neural network migration model based on the deep mixing generation model based on background. The image quality is improved through image processing. The deep hybrid generation model mainly relies on to combine confrontation network generation and self-encoder. In this paper, unsupervised and supervised image style migrations are designed according to the different basic tasks of image style migration. On this basis, unsupervised image style migration of combative neural networks based on cyclic consistency and supervised image style migration of adversarial networks based on cross-domain self-encoder are proposed. This paper further improves the quality of created images by introducing an unsupervised and supervised image style migration standard dataset.