Figure 1: Tactile-Texture-Transfer Pipeline. Users create textures with day-to-day items, which are transferred to their chosen images using Neural Style Transfer Algorithms

​Neural Style Transfer uses deep neural networks to transfer the texture of one image onto another image. What if we could transfer to images textures we could touch and feel as well, right then and there? Tactile-texture-transfer is an experiment that aims to bridge that gap between tangible textures and the immaterial digital world of images. Users can create their own textures on a surface using everyday things - for example, sand, pebbles, and beans. An image of this surface is captured by a phone camera mounted over the surface. This tactile texture that the user has created can then be transferred onto an image of their choosing by using neural style transfer algorithms.

The figure to the right shows an example of this. The user creates three textures from different food items. They then use their phone camera to take a picture, which they can immediately transfer to an image of their choice. Combined textures like the one at the bottom of Figure 2 are also possible, and lead to some interesting results!​

This could have some interesting applications - for example, for transferring live textures to objects in mixed-reality.

Gif: Texture of Garbanzo Beans being transferred to image, epoch 1 to 200

Figure 2: Example tangible textures created by users being transferred to an image