Multiple Style Transfer Emphasizing Visual Saliency

Abstract

An image style transfer is known as an artistic work with computer vision. Conventional approaches in the style transfer have focused on texture extraction and transfer. These approaches developed a successful style transfer but not the content preservation. Damage occurs due to the style transfer of the entire image. When we try to do the style transfer strongly, the contents which have to be preserved are distorted and spoiled. Here we suggest the multiple styles transfer to preserve the content which we aim to do that. The algorithm contains three main techniques – Style transfer with deep learning, Visual saliency and Semantic segmentation. With these computer vision techniques, we apply the multiple styles into a single image successfully. Our work provides the improvement of visual saliency and the better artistic work with the style transfer. Furthermore, it can contribute new sight with visual saliency in philosophy.