
It is well known that all images of the surface (from a fixed view point) under varying illuminations span a 3D space (or illumination cone) defined by the three images (axes of the cone) b1, b2, b3 [1]. With the texton model by wavelet bases, we can decompose the axes images b1, b2, b3 into a triplet of coupled 2D textons, which is defined as "lighton". Example 1:input images under varying illuminations Three image bases computed by SVD from the above input images.
Example 2:input images under varying illuminations Three image bases computed by SVD from the above input images.
References:
