Image Style Transformation Based On Deep Learning

Deep learning is undoubtedly one of the most attractive research areas.Various methods based on the deep learning network usually have excellent performances in a large number of artificial intelligence research and application scenarios.The image recognition technology based on convolutional neural network is a very typical example.As an emerging problem,the image style transformation should be able to make use of the deep learning network.The main research work of this paper is as follows:(1)Inspired by image recognition based on convolutional neural network,a method that has been implemented for realizing image style transformation based on a convolution neural network trained to image recognition is giving:At first,we separate the content information and the style information of images by feature extraction through the network.Then,in order to combine the content information with the style information from different images to generate a new image,we define a loss function for minimization in the network using some optimization algorithm.The new image is regarded as the result of image style transformation.(2)Optimization and improvement of the above methods: We first solve the problem of feature level selection and feature ratio parameter setting in the above methods.Then,we introduces a new method which can be regarded as an improved version of the above method by adding a residual convolutional network to the network and a new definition of loss function.The new method greatly improves the efficiency of image style transformation and realizing fast image style transformation.