Academic Commons

Reports

Surface Orientation and Segmentation from Perspective Views of Parallel-Line Textures

Moerdler, Mark L.; Kender, John R.

This paper describes a particular shape-from-texture algorithm that constrains and defines surface orientations with little a priori knowledge. It has been found to be robust under a variety of conditions. The method uses the change in the spacing of parallel surface markings to derive the orientation and shape of multiple surfaces in synthetic noisy scenes. We first describe the problem domain and the representational approach. Next we outline the mathematical basis of the method, and its straightforward graphical interpretation. Third, we discuss the implementation methodology employed and some concrete implementational issues. We explain the algorithm's response to a number of images, in which various forms of noise or surface perturbation are handled: texel loss or distortion, and multiple or overlapping surfaces. We discuss the relationship between the various forms of noise and the quality of the results. Fourth, we demonstrate preliminary findings of texture-driven segmentation in which this method not only segments adjacent surfaces but also separates two overlapping "transparent" surfaces. We conclude with our future research plans.

Subjects

Files

More About This Work

Academic Units
Computer Science
Publisher
Department of Computer Science, Columbia University
Series
Columbia University Computer Science Technical Reports, CUCS-159-85
Published Here
October 31, 2011