Home

shapefrom

Shape from is a term used in computer vision and graphics to describe a family of methods for reconstructing the three-dimensional shape of a scene or object from two-dimensional observations. The common goal is to infer geometry that is not directly visible in a single image by exploiting cues such as shading, texture, silhouettes, motion, or combinations of these. The umbrella term covers several subfields, including shape from shading, shape from texture, shape from silhouettes (visual hull), shape from motion (structure from motion), as well as related approaches in stereo and photometric methods.

Shape from shading derives surface orientation from variations in image brightness caused by lighting and surface

Shape from texture uses patterns on a surface to infer its orientation and depth from perspective-induced distortions

Shape from silhouettes, or visual hull reconstruction, uses object outlines from multiple views to generate a

Shape from motion leverages parallax across several images to estimate camera motion and reconstruct 3D structure.

Applications include robotics, cultural heritage digitization, 3D modeling, and medical imaging. Limitations include assumptions about lighting,

curvature.
With
a
reflectance
model
(often
Lambertian)
and
known
lighting,
local
surface
normals
can
be
estimated
and
integrated
to
recover
depth.
This
approach
is
intrinsically
ill-posed
and
sensitive
to
non-Lambertian
materials,
shadows,
and
lighting
uncertainty.
or
texture
scaling.
It
requires
a
detectable,
regular
texture
and
some
knowledge
of
the
camera
parameters;
otherwise,
depth
estimates
can
be
ambiguous.
volume
that
is
consistent
with
all
silhouettes.
It
yields
a
coarse
approximation
of
shape
and
cannot
reveal
internal
structure
or
concavities
not
visible
in
any
silhouette.
It
typically
produces
a
sparse
or
dense
3D
model
up
to
a
scale
and
is
robust
to
lighting
changes
but
depends
on
accurate
feature
matching
and
motion
estimation.
texture,
and
object
rigidity,
plus
sensitivity
to
noise
and
occlusions.