没有合适的资源?快使用搜索试试~ 我知道了~
Shape from Shading By Horn
需积分: 27 13 下载量 34 浏览量
2015-02-02
12:15:47
上传
评论 1
收藏 2.29MB PDF 举报
温馨提示
试读
42页
Shape from shading is a simple method to recover surface topography by using a single iamge. This is a original introduction of shape from shading technology.
资源推荐
资源详情
资源评论
4.1
INTRODUCTION
4.1.1
Shading
as
a
Monocular
Depth
Cue
An
image
of
a
smooth
object
known
to
have
a
uniform
surface
will
exhibit
gradations
of
reflected
light
intensity
which
can
be
used
to
determine
its
shape.
This
is
not
obvious
since
at
each
point
in
the
image
we
know
only
the
reflectivity
at
the
corresponding
object
point.
For
some
points
(called
singular
points
here)
the
reflectivity
does
uniquely
determine
the
local
normal,
but
for
almost
all
points
it
does
not.
Consequently,
the
shape
of
the
surface
cannot
be
found
by
local
operations
alone.
For
many
surfaces
the
fraction
of
the
incident
light
which
is
scattered
in
a
given
direction
is
a
smooth
function
of
the
angles
involved.
It
is
convenient
to
think
of
the
situation
as
depending
on
the
three
angles
shown
in
Fig.
4.1:
the
incident
angle
between
the
local
normal
and
the
incident
ray,
the
emittance
angle
between
the
local
normal
and
the
emitted
ray,
and
the
phase
angle
between
the
incident
and
the
emitted
rays.
115
116
The
Psychology
of
Computer
Vision
>
c
Fig.
4.1
Definition
of
the
inci-
dent
(i),
emittance
(e),
and
phase
angle
(g).
'
We
will
show
that
the
shape
can
be
obtained
from
the
shading
if
we
know
the
reflectivity
function
and
the
position
of
the
light
sources.
The
reflectivity
and
the
gradient
of
the
surface
are
related
by
a
nonlinear
first-order
partial
differential
equation
in
two
unknowns.
The
recipe
for
solving
this
equation
involves
setting
up
an
equivalent
set
of
five
ordinary
differential
equations,
three
for
the
coordinates
and
two
for
the
components
of
the
gradient.
These
five
equations
can
then
be
integrated
numerically
along
certain
curved
paths.
For
while
we
cannot
determine
the
gradient
locally,
we
can,
roughly
speaking,
determine
its
component
in
one
special
direction.
Then
taking
a
small
step
in
this
direction,
we
can
repeat
the
process.
The
curve
traced
out
on
the
object
in
this
manner
is
called
a
characteristic.
Figure
4.2
shows
the
characteristics
determined
for
an
experimental
sphere.
Their
projections
on
the
image
plane
will
be
referred
to
as
the
base
characteristics.
The
shape
of
the
visible
surface
of
the
object
is
thus
given
as
a
sequence
of
coordinates
on
characteristics
along
its
surface.
Figures
4.3
and
4.4
show
stereo
pairs
for
three
test
cases.
Figure
4.5
gives
contour
maps
for
the
same
three
objects.
Fig.
4.2
Image
of
a
sphere
and
a
stereo
pair
of
the
characteristic
curves
obtained
from
the
shading.
Fig.
4.3
Stereo
pairs
of
solutions
for
dish-shaped,
spherical,
and
bullet-shaped
objects.
Fig.
4.4
Stereo
pairs
of
same
solutions
as
in
previous
figures
^5
(rotated
90°).
118
The
Psychology
of
Computer
Vision
(a)
(b)
(c)
Fig.
4.5
Contour
maps
of
same
solutions
as
in
previous
figures.
An
initial
known
curve
on
the
object
is
needed
to
start
the
solution.
Such
a
curve
can
usually
be
constructed
near
the
singular
points
mentioned
earlier
using
the
known
local
normal.
The
only
additional
information
needed
is
the
distance
to
the
singular
point
and
whether
the
surface
is
convex
or
concave
with
respect
to
the
observer
at
this
point—such
ambiguities
arise
in
several
other
instances
in
the
process
of
solution
as
will
be
seen.
To
solve
the
equations,
the
reflectivity
as
a
function
of
the
three
angles
must
be
known
as
well
as
the
geometry
relating
light
source,
object,
and
observer.
Multiple
or
extended
light
sources
increase
the
complexity
of
the
solution
algorithm
presented.
But
all
of
this
initially
needed
information
can
be
deduced
from
the
image
if
a
calibration
object
of
known
shape
is
present
in
the
same
image.
Furthermore,
incorrect
assumptions
about
the
reflectivity
function
and
the
position
of
the
light
sources
can
lead
to
inconsistencies
in
the
solution
and
it
may
be
possible
to
utilize
this
information
in
the
absence
of
a
calibration
object.
In
practice
it
is
found
that
if
the
object
is
at
all
complex,
its
image
will
be
segmented
by
edges.
Some
of
these
represent
the
occlusion
of
one
surface
by
another
while
others
are
angular
edges
(also
called
joints
here)
on
a
single
object.
Another
kind
of
edge
is
the
ambiguity
edge.
This
is
an
edge
which
the
characteristics
cannot
cross,
indicating
an
ambiguity
which
cannot
be
resolved
Obtaining
Shape
from
Shading
Information
119
locally.
One
can
solve
inside
each
region
bounded
by
these
various
edges,
but
some
global
or
external
knowledge
is
needed
to
match
up
the
regions.
In
the
case
of
an
angular
edge
on
an
object
one
can
integrate
up
to
the
edge
and
then
use
the
known
location
of
the
edge
as
an
initial
curve
for
another
region.
A
very
similar
situation
obtains
when
one
bridges
a
shadow.
Since
one
edge
of
the
shadow
and
the
position
of
the
light
source
is
known,
we
can
trace
along
the
rays
grazing
the
edge
until
the
corresponding
image
points
fall
on
an
illuminated
region.
Since
we
know
the
path
of
each
ray,
we
can
calculate
the
coordinates
of
the
point
where
it
impinges
on
the
object
by
triangulation.
The
edge
of
the
shadow
(which
need
not
be
on
the
same
object)
can
then
serve
as
an
initial
curve
from
which
to
continue
the
solution.
4.1.2
Applications
A
number
of
interesting
applications
of
this
method
can
be
mentioned.
The
first
of
these
concerns
the
scanning
electron
microscope
which
produces
images
which
are
particularly
easy
to
interpret
because
the
intensity
recorded
is
a
function
of
the
slope
of
the
object
at
that
point
and
is
thus
a
form
of
shading.
In
optical
and
transmission
electron
microscopes,
intensities
depend
instead
on
thickness
and
optical
or
electron
density.
The
geometry
of
the
scanning
electron
microscope
allows
several
simplifications
in
the
algorithm
for
determining
shape
from
shading.
Because
of
the
random
access
capability
of
the
microscope
beam,
it
should
be
easy
and
useful
to
combine
it
with
a
small
computer
to
obtain
three-dimensional
information
directly.
Another
interesting
demonstration
lies
in
the
determination
of
lunar
topography.
Here
the
special
reflectivity
function
of
the
material
in
the
maria
of
the
moon
allows
a
very
great
simplification
of
the
equations
used
in
the
shape-from-shading
algorithm.
The
equations
in
fact
reduce
to
one
integral
which
must
be
evaluated
along
each
of
a
family
of
predetermined
straight
lines
in
the
image.
So
far
we
have
assumed
that
the
surface
is
uniform
in
its
photometric
properties.
Any
nonuniformity
will
cause
this
algorithm
to
determine
an
incorrect
shape.
This
is
one
of
the
uses
of
facial
makeup,
because
by
darkening
certain
slopes
those
slopes
can
be
made
to
appear
steeper.
In
some
cases
surface-markings
can
be
detected
if
they
lead
to
discontinuities
of
the
calculated
shape.
Judging
by
our
wide
use
of
monocular
pictures
(photographs
or
even
paintings
and
woodcuts)
of
people
and
other
smooth
objects,
humans
are
good
at
interpreting
shading
information.
The
shortcomings
of
our
method
which
are
related
to
the
shading
information
available
can
be
expected
to
be
found
in
human
visual
perception
too.
It
will
of
course
be
difficult
to
decide
whether
the
visual
system
actually
determines
the
shape
quantitatively
or
whether
it
uses
the
shading
information
in
a
very
qualitative
way
only.
剩余41页未读,继续阅读
资源评论
houzhuding
- 粉丝: 0
- 资源: 3
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功