Art of Style Transfer Using Convolutional Neural Network: A Deep Learning Approach

Abstract

In the present scenario, many difficulties are being faced while rendering the images with different styles. This makes the image analysis a bit difficult process. The major limiting factor, in this case, is the lack of image representation that explicitly represents the semantic information, which complicates the representation of the image in a way that could further be used to separate the image’s content part from the whole image. To resolve this issue, image representation derived from convolution neural networks (CNN) is used. In the current study, there is an advantage that CNN is optimized for object recognition, which enables it to make high-level image information explicit. This opens up the possibility for the application of a neural algorithm in artistic style which can separate and then recombine the image’s content and its style. This enables us to combine information of random images with well-defined artworks and give it an artistic look that can further be used for different purposes. This deep analysis of images and its information is done by using CNN, and it portrays their capability of high-level image synthesis and image manipulation. The intermediate results in CNN can also be used for feature and content extraction from images.