Generation of Texture: A Case Study with Steel Microstructure Images

Abstract

A lot of work has been done on texture generation techniques. Deep learning based image generation techniques have been extremely successful in generating realistic images. Moreover, reaction-di usion systems have also been successful in generating a wide variety of textures. However, the reaction-di usion systems have never been incorporated in modern deep learning architectures. On the other hand, although a wide variety of images have been generated using traditional computer vision algorithms and deep learning models, very little work has been done on generating the microstructures that are found in abundance in nature. We have explored two established texture generation algorithms for generating steel microstructure images: PatchMatch and DCGAN. We have also tried to combine the reaction-di usion systems with deep learning architectures and have explored the possibility of its success in generating the steel microstructure images.