Application Of Image And Ink Style Rendering Based On Deep Learning

With the rapid development of information technology and biotechnology,also with a substantial increase in computer's compute performance,one of the Computer Version technology named as deep learning has developed very fast in the past four years.Compared to machine learning study before,Deep Learning mainly has more hidden layers and connections between neurons and are more extensive.Because of these reasons,it has been widely used in text,image,voice recognition.In the field of painting,humans have mastered the unique painting skills,this can be called a clever fusion of content and style.The computer does not possess human visual nervous system,so it has been considered not suitable for the creation of art for so long.At present,German scholars have proposed the use of deep convolution neural network(CNN)model for oil painting style by extracting the texture features.These texture features can be fused on any content digital image and generate a picture with both the digital content and the oil painting style.This paper attempts to apply the art style algorithm of convolutional neural network to the creation of Chinese ink painting compared to the previous works and render any picture to Chinese ink style by internet web server.The software architecture is composed of the B/S mode to separate the logical and presentation layers.Users can run the ink style rendering program through the browser and select the style picture,the content picture,adjust the generated image size.The main work of this paper is as follows:1.Present the most popular deep learning framework and convolutional neural network model,analysis of the advantages and disadvantages of the deep learning framework and the convolution neural network model structure,Transplant the VGG,IIIustration2Vec model under Caffe architecture to Tensorflow architecture.2 Write the python program of ink style with Tensorflow.Compared of the Picture quality with different models and the same model produces a combination of different network layers.The gradient of loss function is visualized with Tensorboard in the analysis process.3.Build a server-side program based on Node.js.Using Express as the Web side framework.The front page is done via HTML + CSS + JavaScript.