Estimating vanishing points using visual spatial frequencies of textures on planar surfaces

We present a method for estimating vanishing points of the object planes using the continuous change of visual spatial frequencies of textures on them. Considering that treating 3D textures in real images is a tough problem, we restrict our problem so that at first, human users point out the rough center of the 3D textured region on a planar surface, and its tilt angle has already been estimated. We show that by adequately choosing the direction, of spatial frequency estimation, the change of visual spatial frequencies of textures on planar surfaces is expressed as an analytic curve which has only two parameters. In order to estimate these parameters, which leads to estimating vanishing points, we adopt a voting strategy similar to the Hough transform. Experimental results show that the proposed method can estimate vanishing points well even in real images.