Research Article
The Research and Implementation of Image Style Conversion Algorithm Based on Deep Convolutional Neural Network
@INPROCEEDINGS{10.1007/978-3-030-62483-5_36, author={Huang Yaoqun and Xia Hongyang and Kang Hui}, title={The Research and Implementation of Image Style Conversion Algorithm Based on Deep Convolutional Neural Network}, proceedings={Green Energy and Networking. 7th EAI International Conference, GreeNets 2020, Harbin, China, June 27-28, 2020, Proceedings}, proceedings_a={GREENETS}, year={2020}, month={11}, keywords={Image style conversion Deep learning VGG-19 convolutional network model}, doi={10.1007/978-3-030-62483-5_36} }- Huang Yaoqun
Xia Hongyang
Kang Hui
Year: 2020
The Research and Implementation of Image Style Conversion Algorithm Based on Deep Convolutional Neural Network
GREENETS
Springer
DOI: 10.1007/978-3-030-62483-5_36
Abstract
With the development of deep learning and image technology, the deep convolutional neural network has been widely used to deal with image problems. In this paper, the pretrained vgg-19 convolutional network model is adopted to extract and define the loss function according to the image characteristics, and preset model parameters. The model training is completed through reverse propagation gradient descent and optimization iteration, finally, the artistic painting style conversion of photos is realized. At the same time, by adjusting the size of style weight and content weight, the output image is more inclined to the style picture, or more inclined to the content picture. Finally, the objective evaluation of the output image is has been completed by comparing the image style conversion results of the TensorFlow and the PyTorch. The results show that under the same iteration times, the PyTorch framework has relatively small computation, fast processing speed, and better image color retention effect, in contrast the TensorFlow frame retains more features of style images.
