Title: Image style transfer using convolutional neural networks based on transfer learning

Authors: Varun Gupta; Rajat Sadana; Swastikaa Moudgil

Addresses: Department of Computer Science and Engineering, Chandigarh College of Engineering and Technology, Chandigarh, India ' Department of Computer Science and Engineering, Chandigarh College of Engineering and Technology, Chandigarh, India ' Department of Computer Science and Engineering, Chandigarh College of Engineering and Technology, Chandigarh, India

Abstract: The purpose of an image style transfer system is to extract the semantic image content from the target image and then using a texture transfer procedure display the semantic content of target image in the style of the source image. The uphill task in this context is to render the semantic content of an image but with the advent of convolutional neural networks, image representations have been made much more explicit. In this work, we explore the method for image style transfer using transfer learning from pre-trained models of convolutional neural networks (CNN). Use of these models gives us the power to produce images of a high perceptual quality that are a union of the content of an arbitrary image and the appearance of renowned artworks. Further, this paper compares pre-trained CNN models for image style transfer task and highlights the potential of CNN to deliver appealing images using modern manipulation techniques.

Keywords: image style transfer; convolutional neural networks; CNN; transfer learning; deep learning; machine learning; artificial intelligence.

DOI: 10.1504/IJCSYSE.2019.098418

International Journal of Computational Systems Engineering, 2019 Vol.5 No.1, pp.53 - 60

Available online: 06 Mar 2019 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article