A simple algorithm for shape from shading

Abstract

In this paper we describe a simple shape from shading algorithm which recovers depth from a brightness image, typically in fewer than ten iterations. This algorithm, which is a simplification of the algorithm of Oliensis and Dupuis, is based on a minimum downhill principle which guarantees continuous surfaces and stable results. The algorithm is applicable to a broad variety of objects and reflectance maps.

Needle map recovery using robust regularizers

“…Here we consider the formulation of Brooks and Horn [17], which is couched in terms of unit surface normals. Whilst there are other, more recent and in some cases more elegant, approaches to the problem (for example, [13,14,19]), the Horn and Brooks formulation provides a simple and easily adaptable platform upon which to test the utility of the methods described here. We fully expect that the robust techniques presented in this article may be applied to any scheme which utilizes smoothness or other constraints to achieve a corresponding performance improvement.…”

Shape from Shading: Recognizing the Mountains through a Global View

“…Previous methods [1,9,17,19,20] perform poorly on these examples. Energy minimization approaches suffer from smoothness constraint and local minima: the result surface is flat globally but bumpy locally (Figure 6(e)).…”

“…Local propagation methods can give reasonable smooth surfaces, but fail to give a correct global shape (Figure 6(f)). For a fair comparison, we have already tuned the parameters for the previous methods [1,17] and take the best results. Further, we choose the top two results from the six methods surveyed in [19].…”

“…Shapes around singular points can be computed exactly (without using smoothness constraint) using propagation methods [1,7,11,13]. However, local shapes are determined up to three types of ambiguity convex/concave/saddle [2,10].…”

A graph spectral approach to shape-from-shading

“…We compare the resulting height estimates with the height data for the original surfaces. We also provide a comparative study using the shape-fromshading height recovery method of Bichsel and Pentland [18] as an alternative to our graph-spectral shape-from-shading algorithm. We have chosen the algorithm of Bichsel and Pentland since, from Zhang et al [7], appears to deliver reasonable results on a wide variety of images.…”