Factoring a specular reflection Field Into added iffuse reflection Fields
This paper presents the first stages in exploring an hypothesis that would appear to support a shape from shading, reconstruction algorithm from time series image sequences. The hypothesis arises from a demonstration that stereo contoured images of a diffuse reflection surface can be used to reconstruct a reasonably accurate three-dimensional contour model of the original surface. When the same approach is applied to a true mirror then the reconstructed surface becomes that of the reflected scene. However, if the mirror is dirty, this reconstruction breaks down giving neither the mirror surface nor the scene surface. If the two reflection components from the dirty mirror, from the scene and from the mirror surface, assuming each to obey Lambert’s Law, can be separated then it should be possible to apply the original simple algorithm to each image factor to define two three dimensional surfaces. This paper explores a way in which a time series sequence of images such as a video or film, or stereo images might be used to factor a specular reflecting surface into an equivalent set of added “Lambertian” surfaces.
KeywordsSpecular Reflection Stereo Image Matte Surface Mirror Surface Reflection Component
Unable to display preview. Download preview PDF.
- 2.D. Vernon, “Removal of superimposed reflections from dynamic image sequences”, in “Irish Machine Vision and Image Processing Conference and Eight Ireland Conference on Artificial Intelligence”. Magee College, University of Ulster. 10–13 September 1997.Google Scholar
- 3.A. L. Thomas, “Speculation on virtual surfaces in machine vision systems”. in “Irish Machine Vision and Image Processing Conference and Eight Ireland Conference on Artificial Intelligence”. Magee College, University of Ulster. 1013 September 1997, pp 55–60Google Scholar
- 5.Olivier Faugeras, “Three-Dimensional Computer Vision ”, The MIT Press, London, 1993Google Scholar
- 6.Richard Hartley and Andrew Andrew, “Multiple View Geometry in Computer Vision ”,2000.Google Scholar