International Conference on Computer and Knowledge Engineering

Home / 11th International Conference on Computer and Knowledge Engineering

WBT-GAN:Wavelet based Generative Adversarial Network for Texture Synthesis

Authors :

Sara Saberi moghadam¹ Reza Azmi² Maral Zarvani³

1- Faculty of Engineering , Alzahra University 2- Faculty of Engineering , Alzahra University 3- Faculty of Engineering , Alzahra University

Keywords :

Image to image translation, Texture synthesis, Wavelet transform, Conditional generative adversarial networks, Fashion AI

Abstract :

Texture transfer and synthesis is one of the most widely used topics in image-to-image translation, an important field in computer graphics. With the development of Generative Adversarial Networks(GANs) and their variants, they have become the state-art-of-the-art method of this field. One of the most important challenges for these networks is to generate high-resolution and realistic images that depend on high-frequency features such as textures and edges. One of the most powerful time-frequency transformations in image processing is the Wavelet Transform(WT), which has the ability to describe images at multi-level resolution. In this study, we propose a generative model called WBT-GAN, which is an extension of the existing network Texture-GAN, by using the four-level WT and defining the loss function based on it in the objective function. Experimental results show that these changes have led to improved image resolution and sharpening and better texture spread.