This paper presents a new approach to extract information of 3D-object shapes, by analysis of the 2D-projection of a textured object surface. This problem is well known as “Shape from Texture”. The proposed approach employs a M-channel wavelet filterbank for feature extraction. Analytical expressions have been derived for the imaging geometry, in polar coordinate system, for an inclined surface with sine wave patterns. The proposed algorithm has been evaluated on real world image data, and it is shown that this approach can also be successfully applied for 3D corrugated textured surfaces as well. Further, it has been shown that the “texture gradient” when used as a measure for the geometric inclination of a texture surface, allows to segment images of 3D-objects and also to detect changes of depth within 3D-objects.
Print ISSN: 0178-2312
Volume: 54, 10/2006
Pages: 475 - 485