What if materials or surface textures used in architecture were only renderings of the material’s data-space? This ongoing project understands texture samples as pure data vectors and aims at mapping them in such a way that one could easily tell the distance between oak and birch and render the inbetween as some sort of meta-material.
Input of the following explorations are the 54 pictures of wood assembled in this picture:
Original image taken from here.
Wood was chosen as an input because of its big variety of colors and textural features on the one hand, while on the other hand remaining self-similar to a certain degree. Other applications of the methods in use are for example face-recognition or the related Eigenfaces. There also, the computer learns what “faceness” is (roughly speaking to dark spots at mid height on either side plus another one in the middle further down) by being trained with a series of sample face images. Every new image of a face can then be compared against the principle components of the training data set and expressed as a product of these eigenvectors with different weights.