Shape from texture: General principle (Q1122365)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Shape from texture: General principle
scientific article

    Statements

    Shape from texture: General principle (English)
    0 references
    0 references
    0 references
    1989
    0 references
    Surfaces on which the ``texture'' consists either of a finite set of isolated points or a finite set of isolated line segments are inspected. Such a surface is projected onto an image plane. It is assumed that the projections of isolated points are isolated points and that these projections can be identified in the image (similarly for line segments). The authors claim that if on the original surface the texture density of the points (line segments) is homogeneous (in a sense given by them), then one can recover information regarding the shape of the surface from the image. No assumptions besides homogeneity are placed on the texture. Furthermore, the proposed approach is strictly mathematical/computational, no heuristics or ``high intelligence'' is required. (One wonders: why is this article in the journal Artificial Intelligence?) The basic idea is to think of the observed texture as if it were a distribution, that is a linear functional operating on a set of test functions defined on the image plane. For example, the result of applying a point texture to a test function m is the sum of the values of m at the points of the texture. It is assumed that (for the m's that we might want to use) this sum can be computed. Alternatively, if the surface is parameterized, the value of the sum can be expressed as a function of the parameters, except for a multiplicative constant which depends on the unknown texture density. This constant can be eliminated by taking a ratio of two sums for two different m's. Each such ratio gives us a nonlinear equation in the unknown parameters; by choosing a sufficiently large number of m's, one can hopefully solve for the parameters based on the observed images. The approach is discussed in detail for perspective imaging of planes and for orthographic imaging of quadric surfaces. Iterative numerical schemes for solving the nonlinear equations are proposed and illustrated on examples using computer simulation. However, no physical experiments are reported and there is no discussion of the stability of the iterative procedures under the unavoidably inexact conditions that exist in the real world.
    0 references
    shape
    0 references
    texture
    0 references
    differential geometry
    0 references
    computer vision
    0 references

    Identifiers