Estimating the Affine Transformation between Textures

A method is presented that solves the affine transformation between two images of the same texture. This problem is encountered in many areas of Computer Vision such as object retexturing and it is an essential part of many Shape-from-Texture algorithms. The new method is based on transforming each image such that the texture is isotropic, then solving the similarity transformation between the isotropic versions. We point out the difficulties in using other methods such as feature point methods. Quantitative results are given for images created with the Brodatz textures, followed by results for real images. This is followed by an application of the method: placing an object on the textured surface.