|
|
|
“Direct Shape Recovery from Texture Information”
by Kyoung Mu Lee and C.-C. Jay Kuo
January 1993
Most existing shape from texture algorithms extract the surface orientations of planer objects from projected texture variation in an image. In this research, we propose a new shape from texture (SFT) algorithm which recovers the surface heights of curved objects directly. By examining the perspective transformation of a local planar surface patch onto the image plane, we establish a nonlinear relationship between the measured local density of texture primitives on the image plane and the orientation and position of the surface patch in the world space. Then, by approximating a curved surface with a union of triangular surface patches and dividing an image into a set of nonoverlapping triangular domains, we can relate the densities of textural primitives in triangular domains to nodal height variables of the curved surface. This procedure leads to a nonlinear system of equations which are then solved via a successive linearization scheme. Estimation of the local textural density using dots and line segment edges of texels is also discussed. Experimental results are given to illustrate the performance of the new algorithm.