Distortion
Distortion
by
B. Hönlinger and H. H. Nasse
Carl Zeiss
Camera Lens Division
October 2009
Preface
“I found somebody say ‘terrible moustache
distortion’, but I don’t see any distortion at all”.
As you can see in above statement in a forum discussion one
can have very different opinions about distortion of lenses. For
some photographers straight lines are of low importance or their
subjects don’t have them. Others strive for perfection and spend
a lot of efforts, to heal geometric aberrations at least in the
image post-processing.
Those who try to avoid these efforts search for lenses which
deliver perfect quality right from the beginning without later
improvements. At this point the problem with numbers arises,
because absolute perfection exists only in a few types. Most
lenses exhibit at least small distortion errors, and to decide
whether they are acceptable, one has to understand the
numbers in data sheets and in lens test publications. This issue
of Camera Lens News tries to support you a little bit in
understanding this matter.
And since distortion is of major importance in wide-angle lenses,
we will in addition deal with some other strange effects typical for
this class of lenses, so that you understand why image
composition with the short ones is at the same time so difficult
and attractive.
Carl Zeiss
Camera Lens Division
October 2009
What is distortion?
The majority of camera lenses produce
images in line with the laws of central
perspective. This kind of projection of threedimensional space onto a two-dimensional
image surface is also called gnomonic
=
projection. This Greek term (
gnomon = shadow-producing rod) denotes a
type of sundial because, as with a sundial, an
image point is produced by connecting a
point in the object space to the center of the
projection using a straight line; the image
point is where this straight line intersects
the flat projection surface.
In painting, central perspective was
invented during the time of Renaissance.
The methods used by the artist to strictly
apply this perspective are shown to us by
Albrecht Dürer:
Instructions in the application of perspective, woodcarving by Albrecht Dürer,
approx. 1527
The center of projection of the image is the
tip of the rod used by Dürer's artist to draw
attention to the charms of the young lady
in this picture. The projection surface is
the frame fitted with a grid placed between
the two, and it is used by the artist to
transfer the image points onto his drawing
paper with the proper perspective.
The center of projection of camera lenses
is their entrance pupil, i.e. the image of
the aperture stop viewed from the front.
When taking panorama photos, it is
necessary to swivel around the entrance
pupil to ensure that objects in the
foreground and background are not shifted
with respect to each other.
This special point is often also called the
nodal point, but this term has a very
different meaning in optics. There is
nothing particularly mysterious about the
entrance pupil either, since anyone can
see it without any special aids and
estimate its approximate position.
Carl Zeiss
However, the entrance pupil is not the
physical aperture stop, but rather its virtual
image – and as such may even be
situated outside the lens altogether. This
is often the case with short telephoto
lenses.
The gnomonic (central perspective)
projection has the special feature where
all straight lines of the object space are
reproduced in the image as a straight line
again regardless of where they are
situated and to where they are projected.
Lenses with this property are called
‘rectilinear’. During the history of
photographic lenses they appeared first in
th
the sixties of the 19 century.
Distortion is defined as a lens aberration in
which this property is no longer exactly
fulfilled. A lens that exhibits distortion
produces slightly curved images of all
those lines that do not pass through the
center of the image. Thus this effect is
also called ‘curvilinear distortion’.
Camera Lens Division
3 / 28
What can we learn from the data sheet of the lens?
The reason for the curved images of
straight lines is that the image scale is not
constant throughout the entire image field.
In other words: the focal length of a lens
showing distortion changes with the
distance of an image point from the optical
axis.
If the reader allows me to use the
language of mathematics, it will be much
easier to comprehend what the numbers
we use to describe distortion actually
mean.
Let us first look at the image equation of a
rectilinear lens with ideal gnomonic
projection:
u'
f ' tan W
W is the object-side field angle, i.e. the
angle between the optical axis and the line
from the object point to the entrance pupil,
u’ is the off-axis distance, i.e. the distance
of the image point from the optical axis. To
express this equation in words: the off-axis
distance is proportional to the tangent of
the field angle, and the focal length f’ is
the proportionality constant.
I would now like to mention that this is
exactly how the focal length of lenses is
defined and measured: one measures the
angle to the optical axis and the off-axis
distance of the corresponding image point
created by a ray of light arriving from a
source at an infinite distance. You may
also encounter measurements, in which
the focal length is calculated in the near
range from the distance and image scale
factor. However, this result often deviates
substantially from the specifications of the
manufacturer.
However, this does not necessarily mean
that the manufacturer has cheated vastly.
Many lenses change their focal length
when focused on short distances, and
calculations of this kind fail to take into
account other lens parameters such as
principal plane distances. For this reason,
a long zoom lens, for example, often has
smaller object fields at short distances
than a comparable prime lens.
* In optical tradition the off-axis distance of an image
point I also called ‘image height’. This should not be
confused with the vertical frame length.
Carl Zeiss
Another factor must be included in the
gnomonic image equation if the focal
length is not constant throughout the field
of view:
u'
f' 1
D%
100
tan W
D is the measure of the distortion error. If
D=0, the term in parentheses is equal to 1,
and the situation is the same as with the
ideal lens.
If D is a positive number, the term in
parentheses is larger than 1. The off-axis
distance u’ is then larger than with the
distortion-free lens of the same focal
length. Since the distortion increases with
increasing distance in most cases, a
rectangle is distorted to a pincushion
shape. We shall see later though that
pincushion-shaped images can also occur
with negative values of D.
If D is a negative number, the term in
parentheses is smaller than 1. The off-axis
distance u’ is then smaller than with the
distortion-free lens of the same focal
length. The lines of a rectangle are
bulging and we call this a barrel shape.
In both cases considered here, the image
point is shifted in a radial direction, i.e. on
the radius of the field of view. Accordingly,
parameter D is also called radial
distortion. Our data sheets specify the
radial distortion by means of a curve, i.e.
as a percentage value as a function of the
off-axis distance *.
This curve contains all information about
the geometric properties of the lens. For
proper interpretation, though, we need to
understand it even better. What makes our
eyes so sensitive to distortion? Taking a
photo of concentric circles, for example a
target for shooting practice, we can barely
recognize distortions as long as they
remain moderate since we have little
feeling for the correct absolute size of a
circle. In contrast, we can see pretty well
that straight lines are reproduced with
some curvature, i.e. we need to
understand the relationship between radial
distortion and this kind of bending of
straight lines.
Camera Lens Division
4 / 28
Radial distortion and TV distortion
Types of distortion
1. Pincushion distortion
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
0
4
8
12
16
Radial distortion [mm]-
Radial distortion [%]-
Pincushion distortion at an already detectable
level in a lens for the 35mm format is
described by the following curve:
20
Off-axis distance [mm]
The black curve shows that the radial
distortion error increases gradually from zero
in the middle of the image to 3% in the corner
of the image where the focal length is 3%
larger than in the middle.
The red curve shows the absolute magnitude
of the radial shift of the image points in units
of millimeters. These values tell us that
distortion shift is more than ten times larger
than the usual circle of confusion. Thus
perceivable distortion is nearly never caused
by poor build quality or shock damage of the
lens. Distortion is not an issue for repair
service, it is determined by the lens design.
And distortion is not changed by stopping
down.
Looking at this curve, one keeps in mind:
"The lens exhibits a 3% distortion", but might
then be surprised if a test report claims: "The
lens shows 1.1% pincushion distortion". How
can we reconcile this apparent discrepancy?
Well, the test report does not refer to radial
distortion, but rather to TV distortion. This is
a measure of how strongly the image of a
straight line that is situated at the edge of the
image (in particular on the long edge of the
image in a rectangular format) is curved. This
amplitude of curvature is then related to the
total frame height and expressed as a
percentage value.
It is therefore very important to note which
percentage value is actually meant. The
values of TV distortion are always smaller
than the radial distortion specified by us.
Carl Zeiss
DTV
H
100
H
Sometimes the total height difference of
the distorted rectangle is related to the
frame height. This is then twice the above
figure.
In order to understand how these fairly
different numbers arise, let us look at a
line at the upper edge of the 35mm format.
A point in the middle of this line has a
distance of 12 mm from the middle of the
image. Here, the lens shows approx. one
percent radial distortion. Consequently,
the image of this line is not only curved,
but also shifted by 0.12 mm. However, we
do not notice this.
If we travel along the line until we get to
the corner, the off-axis distance increases
from 12 to 21.6 mm and, according to the
curve presented above, the radial
distortion increases to 3.1% in the
process. The endpoint of the line is
therefore shifted in a radial direction by
0.031 x 21.6 = 0.67 mm. However, in the
24x36 rectangle of the 35 mm format, the
diagonal is inclined by 56° with respect to
the vertical direction. Splitting the radial
distortion into horizontal and vertical
components, the distortion shift in the
vertical direction is 0.67 x cosine(56°) =
0.37 mm. Accordingly, the vertical shift of
the image points in the middle of the line
and at its end is 0.12 mm and 0.37 mm,
respectively. The amplitude H is the
difference between these two values, i.e.
0.25 mm. This corresponds to approx.
1.1% for a frame height of 24 mm.
Camera Lens Division
5 / 28
TV-Distortion at long edge = 1.1 %
This diagram shows a simulated image characterized by the pincushion distortion
described above. The crosses are the image points that are imaged with distortion error.
The thin red lines are reference lines without curvature. For reasons of symmetry, it is
sufficient to show just one quarter of the image, meaning that the middle of the image is
situated in the lower left corner of the graph. The blue lines show the crop factor of 1.5x.
0.4
1.6
0.35
1.4
0.3
1.2
1
0.25
0.2
0.8
0.15
0.6
0.1
0.4
0.05
0.2
TV distortion [%]-
Amplitude of curvature
[mm]
TV distortion at long edge = 1.1 %
0
0
0
5
10
15
20
Line distance from centre [mm]
horizontal, barrel
wave type
wave type
pincushion
pincushion
vertical, barrel
Each point in this graph represents the amplitude of line curvature plotted over the
distance of the line from the image centre. The amplitude of curvature increases uniformly
from the middle towards the edge in this distortion profile. TV distortion values usually
refer to a line on the long edge of the image (symbol filled in blue). Curvature is less
pronounced on the short edge, since this line proceeds only through off-axis distances
between 18 and 21.6 mm.
Carl Zeiss
Camera Lens Division
6 / 28
2. Barrel distortion
A lens with substantial barrel distortion of
straight lines shows negative radial distortion
whose absolute amount increases towards
the corner:
0
4
8
12
16
Let us have the negative radial distortion
drop just a little bit more, let's say to -7% i.e. it is 1.4 times larger than before:
20
0
-1.0
0.0
12
-2.0
-3.0
18
-4.0
22
-5.0
-6.0
Radial distortion [%]-
Radial distortion [%]-
0.0
4
8
12
16
20
12
-1.0
-2.0
18
-3.0
-4.0
-5.0
-6.0
22
-7.0
-8.0
Off-axis distance [mm]
Off-axis distance [mm]
The lens reaches -5% radial distortion in the
corner of the image, which is a typical value
of many wide-angle to tele zoom lenses at
shortest focal length.
The curvature of a peripheral line at the long
edge of the format is easy to calculate. We
need to bear in mind that the radial distortion
in the corner needs to be multiplied by the
cosine of 56° = 0.55 in order to take into
account only its contribution to the shift in the
vertical direction.
DTV = (21.6 x 0.05 x 0.55 – 12 x 0.02)/24
Values like this do indeed occur with reallife lenses when these are used outside
their normal field of application, e.g. a zoom
lens in an extreme macro-photography
setting.
The same calculation for the long edge of
the format as before yields:
DTV = (21.6 x 0.07 x 0.55 – 12 x 0.01)/24
This means approx. 3% TV distortion for
this line - quite a bit. Although we increased
the maximal radial distortion by only 40%,
we managed to double the TV distortion.
Result: 1.5 % TV distortion.
Similarly, we can calculate the curvature of a
peripheral line at the short edge of the format,
but now have to take into account the
inclination of the diagonal with respect to the
horizontal direction. This angle is only 34°
and the cosine of 34° is 0.83.
DTV = (21.6 x 0.05 x 0.83 – 18 x 0.04)/24
The result obtained (0.73%) is smaller again,
as expected.
This shows how easy it is to get lost in the
dense forest of percentage values! Let me
demonstrate this even better by making the
lens just a little bit worse.
Carl Zeiss
The cause of this phenomenon is that we
are considering a difference (subtraction) in
our calculation, to be precise, the difference
between the smallest and the largest shift
due to distortion. In the second lens
considered above, not only did the distortion
increase in the corner but it became smaller
at the off-axis distance of 12 mm at the
same time.
Thus, the perception of distortion, i.e. the
curving of what are actually straight lines,
by our eyes, appears to depend on the
change of the radial distortion along a line.
And, at this point, TV distortion appears to
be the better measure. But I will show you
some examples demonstrating the opposite
in the next chapter - but first let us have a
look at the image simulations of the two
lenses:
Camera Lens Division
7 / 28
TV-Distortion at long edge = 1.5 %
TV-Distortion at long edge = 3 %
Simulated image with the negative radial distortion described on the previous page. Both
lenses exhibit strong barrel distortion in the entire field of view. The doubling of TV
distortion in the lower example results from the higher gradient of radial distortion towards
the edge of the image.
Carl Zeiss
Camera Lens Division
8 / 28
TV-Distortion at long edge = 0.9 %
TV-Distortion at long edge = 0.7 %
However, using the two lenses in the APS-C format, the second one shows only 0.7% TV
distortion and thus is even slightly better. This is because its radial distortion curve is
initially flatter at the smaller image heights.
Carl Zeiss
Camera Lens Division
9 / 28
Not all motifs reveal the pincushion distortion
of a zoom lens in longest focal length setting
as clearly as this one. Horizontal and vertical
lines close to the frame edges enable an
easy comparison.
Even if this door is already quite old all its
elements are straight, parallel and rectangular
to each other. A wide angle zoom makes us
believe that the door is not really closed.
The same zoom lens as above shows a
pronounced barrel type of distortion when it is
used at the shortest focal length. This causes
the illusion that the brick surface forms a kind
of dome towards the observer.
The image of wooden floor covering
boards on the left exhibits a kind of
distortion which seems to be a mixture of
both types mentioned above. Please
read on the following pages how this can
happen.
The image actually was a 2164x256 crop
from a 4288x2848 file taken from the top
right edge of the frame. The height was
then exaggerated by a factor of 8, much
like we will do it in some of the next
graphs.
Carl Zeiss
Camera Lens Division
10 / 28
3. Moustache distortion
4
8
12
16
20
0.8
-1.0
0.6
0.4
-2.0
0.2
-3.0
0.0
-4.0
-0.2
-5.0
-0.4
Slope [% / mm]-
Radial distortion [%]-
0
0.0
Off-axis distance [mm]
The radial distortion of this lens is negative
throughout; it reaches its highest amount of
-2.6% at 16mm off-axis distance and then
decreases to -0.8% towards the corner.
Many lines that are parallel to the long edge
of the format and extend across the entire
width of the image have some points in the
red and some points in the blue section.
Accordingly, there must be a switch
between barrel and pincushion distortion on
these lines. We say that they exhibit
moustache distortion, also called wave-type
or gull-wing distortion.
Looking at the entire image and at lines that
differ in length, all three types of line
curvature are evident. This is shown in an
exaggerated manner in the following
diagrams:
Half lines parallel to the long frame edge
0
Vertical shift of image points
[mm]-
Unfortunately, it is not always this easy to
calculate the TV distortion from the radial
data since there are many lenses whose
radial distortion does not increase or
decrease uniformly between the middle and
the corner but rather reaches an extreme
value somewhere in between. This is caused
by countermeasures in the design to limit
negative distortion. Many retro-focus wideangle lenses have a radial distortion function
like the one shown in the following example:
4
8
12
16
0.00
0.40
-0.05
0.35
-0.10
0.30
-0.15
0.25
-0.20
0.20
-0.25
0.15
-0.30
0.10
-0.35
0.05
-0.40
0.00
Length along horizontal lines [mm]
close to frame edge
Line distance: 8 mm
type 1, pincushion
While we may have remembered until now
the simple rule: "positive D
pincushion,
negative D
barrel“, this lens teaches us
that this simple rule is an oversimplification
(unfortunately it was also said on earlier
ZEISS data sheets).
The pure pincushion type 1 is also plotted
as a red line for purposes of comparison.
Although it has even slightly higher
amplitude than the black line, this may often
be less evident since it is not as steep near
the edge.
Looking at the image of lines that are parallel
to the edge and whose points have off-axis
distances of between 0 and 16 mm, the
distortion curve looks just like that of type 2 barrel distortion is evident.
Looking at lines near the short edge of this
format and whose shortest off-axis distances
exceed 16 mm, the distortion curve of this
section looks just like that of type 1 - it is
negative, but the values increase towards the
corner. Accordingly, we see pincushion
distortion.
The following two lines are parallel to the
short edge of the format and show purely
barrel or pincushion features:
Carl Zeiss
Length along vertical lines [mm]-
This shows that the attributes "pincushion"
and "barrel" actually do not depend on
whether the radial distortion is positive or
negative, but rather on the slope of the
distortion function. This change, in % per mm,
is drawn in color in the diagram above: red
where it is negative and blue where it is
positive.
Half lines parallel to the short frame edge
-0.50
12
-0.40
-0.30
-0.20
-0.10
0.00
10
8
6
4
2
0
Camera Lens Division
Horizontal shift of image points [mm]
close to frame edge
Line distance: 9 mm
11 / 28
TV-Distortion at long edge = 1 %
Simulated image with the negative radial distortion with local extreme value described on
the previous page. Pincushion-type dominates on the edges of the 35 mm format,
whereas the distortion is mainly of barrel-type in APS-C format (blue lines).
0.4
1.6
0.35
1.4
0.3
1.2
1
0.25
0.2
0.8
0.15
0.6
0.1
0.4
0.05
0.2
TV distortion [%]-
Amplitude of curvature
[mm]
TV-Distortion at long edge = 1 %
0
0
0
5
10
15
20
Line distance from centre [mm]
horizontal, barrel
wave type
wave type
pincushion
pincushion
vertical, barrel
The distortion profile of lines with a different distance from the middle of the image
is quite complex: after increasing gradually at first, it decreases for a while until
there is a dramatic increase towards the edge. The shape changes in the
process. The standard TV distortion (blue triangle) does not represent the line
which exhibits the most severe curvature.
Carl Zeiss
Camera Lens Division
12 / 28
The lens discussed above can be used to
take images without annoying distortion if one
ensures that prominent lines of the object that
are near the long edge of the format do not
extend into the extreme corner and by
avoiding the placement of critical lines too
close to the short edge of the format.
This is clearly evident in the strongly
exaggerated plot:
Half lines parallel to the long frame edge
0
4
8
12
16
0.00
Vertical shift of image points
[mm]-
The
example
shown
above
clearly
demonstrates that distortion errors, much like
many other properties of lenses, are difficult
to describe in detail by means of a single
number. The maximum value of radial
distortion can be misleading; the numerical
value of TV distortion is sometimes
somewhat closer to subjective perception but it still does not tell us enough about the
distortion throughout the image and it does
not tell us the profile of distortion along a line.
-0.10
-0.20
-0.30
-0.40
-0.50
Length along horizontal lines [mm]
close to frame edge
Line distance: 8 mm
4
8
12
16
Radial distortion [%]-
0.0
0.0
-3.0
-0.2
-4.0
-5.0
-0.60
-0.50
-0.40
-0.30
-0.20
-0.10
0.00
8
6
4
2
0
Horizontal shift of image points [mm]
0.2
-2.0
-0.70
10
close to frame edge
0.4
-1.0
-0.4
Off-axis distance [mm]
Since the extreme value of radial distortion is
now situated near the corner of the image
and since the values increase only over a
very short distance and much less steeply,
barrel distortion predominates in much of the
image.
For lines at the short edge of the format with
off-axis distances between 18 and 21.6 mm,
the radial distortion is quite large. As a result,
the entire line is shifted more strongly, but
this is not perceived by us. The crucial point
is that the change of the radial distortion in
this range of image heights is very small,
since this causes the lines at this edge of the
image to be practically straight - despite - 4%
radial distortion.
Carl Zeiss
-0.80
12
20
Slope [% / mm]-
0
Length along vertical lines [mm]-
Half lines parallel to the short frame edge
The limited value of having only a single
number is even more evident in the following
example in which the amount of negative
radial distortion is increased to - 4% and the
curvature is nevertheless lower for the most
part:
Line distance: 12 mm
Precisely because the slope of the
distortion curve near the edge of the
image is small and has a less steep profile
than in the previous example, the
distortion, which appears to be high in the
data sheet, is "masked" fairly well. Our
perception is often particularly sensitive at
the periphery of the image since the
straight edge serves as a reference to the
eye, much like a ruler.
The figure of the TV distortion is moderate
but it withholds something from us, namely
that the curvature further inwards is
somewhat larger than on the periphery of
the image. It also does not tell us that the
short edge does not show any
recognizable distortion at all (see next
page).
Camera Lens Division
13 / 28
TV-Distortion at long edge = 0.7 %
Simulated image of the negative radial distortion described on the previous page. Barrel
distortion dominates on the upper edge of the image whereas the vertical edge of the
image on the right is basically free of curvature – in spite of a - 4% radial distortion figure.
0.4
1.6
0.35
1.4
0.3
1.2
1
0.25
0.2
0.8
0.15
0.6
0.1
0.4
0.05
0.2
TV distortion [%]-
Amplitude of curvature
[mm]
TV-Distortion at long edge = 0.7 %
0
0
0
5
10
15
20
Line distance from centre [mm]
horizontal, barrel
wave type
wave type
pincushion
pincushion
vertical, barrel
For lines with different distances from the middle of the image, the profile of
curvature is simpler than in the previous example: it increases gradually, reaches
a maximum value at a distance of between 9 and 10 mm from the middle and
then decreases again and disappears almost completely at the short edge. The
standard TV distortion (blue triangle) does not indicate the largest curvature.
Carl Zeiss
Camera Lens Division
14 / 28
TV-Distortion at long edge = 0.9 %
TV-Distortion at long edge = 0.5 %
12
16
Radial distortion [%]-
20
0
-0.5
-1.0
-0.1
-1.5
-2.0
-0.2
-2.5
-3.0
-0.3
4
8
12
16
0.0
0.0
20
0.1
-0.5
-1.0
0.0
-1.5
-2.0
-0.1
-2.5
-3.0
Off-axis distance [mm]
Slope [% / mm]-
8
Radial distortion [%]-
4
Slope [% / mm]-
0
0.0
-0.2
Off-axis distance [mm]
Two more examples with the same maximum value of negative radial distortion: they
clearly show us that the higher slope of the radial distortion curve causes more curvilinear
distortion at the frame edges (top and left graph).
Carl Zeiss
Camera Lens Division
15 / 28
TV-Distortion at long edge = 0.9 %
TV-Distortion at long edge = 0.9 %
12
16
20
0
0.7
-1.0
0.5
-2.0
0.3
-3.0
0.1
-4.0
-0.1
-5.0
-0.3
4
8
12
16
20
0.0
0.4
-1.0
0.2
-2.0
0.0
-3.0
-0.2
-4.0
-0.4
-5.0
Off-axis distance [mm]
Slope [% / mm]-
8
Radial distortion [%]-
4
Slope [% / mm]-
Radial distortion [%]-
0
0.0
-0.6
Off-axis distance [mm]
These two examples and one on the previous page have all the same amount of TV
distortion. When you however examine the lines close to the frame edges you will notice
quite obvious differences.
Carl Zeiss
Camera Lens Division
16 / 28
Angle of Inclination
The examples we have discussed above
teach us that distortion can hardly be
described by one single number. And the
distortion charts of the data sheet need
further ‘translation’ to get what we see.
A much better and more informative
presentation of distortion features would be a
kind of map which we show on this page to
explain the four most important types of
distortion.
4
8
12
16
20
5.0
0.5
4.0
0.4
3.0
0.3
2.0
0.2
1.0
0.1
0.0
Slope [% / mm]-
Radial distortion [%]-
0
This map represents the inclination angle
of horizontal lines in a quarter of the
image. The middle of the image is again
the lower left corner of the map. Without
distortion this angle would of course be 0°
within the complete image area. A positive
inclination means that a line in the image
rises from left to right in the map (yellow
and red colors). A line with negative
inclination angle drops from left to right.
Each single color represents a range of
0.5°. Where the colors change rapidly
lines are strongly curved.
0.0
Inclination
[°]
Off-axis distance [mm]
4
8
12
16
20
0.0
0.0
-1.0
-0.1
-2.0
-0.2
-3.0
-0.3
-4.0
-0.4
-5.0
Slope [% / mm]-
Radial distortion [%]-
0
-0.5
4
8
12
16
20
0.7
-1.0
0.5
-2.0
0.3
-3.0
0.1
-4.0
-0.1
-5.0
Slope [% / mm]-
Radial distortion [%]-
0
-0.3
4
8
12
16
20
0.5
-1.0
0.3
-2.0
0.1
-3.0
-0.1
-4.0
-0.3
-5.0
Slope [% / mm]-
Radial distortion [%]-
0
-0.5
Off-axis distance [mm]
Carl Zeiss
2-2.5
1.5-2
1-1.5
0.5-1
0-0.5
-0.5-0
-1--0.5
-1.5--1
-2--1.5
-2.5--2
Inclination
[°]
Off-axis distance [mm]
0.0
2-2.5
1.5-2
1-1.5
0.5-1
0-0.5
-0.5-0
-1--0.5
-1.5--1
-2--1.5
-2.5--2
Inclination
[°]
Off-axis distance [mm]
0.0
2-2.5
1.5-2
1-1.5
0.5-1
0-0.5
-0.5-0
-1--0.5
-1.5--1
-2--1.5
-2.5--2
Camera Lens Division
2-2.5
1.5-2
1-1.5
0.5-1
0-0.5
-0.5-0
-1--0.5
-1.5--1
-2--1.5
-2.5--2
Inclination
[°]
17 / 28
Types of lenses
The optical reasons of lens distortion are
related to the distribution of the refractive
power in the lens. However, at the present
time this should only be used for predicting
distortion with great caution since many
modern corrective tools are available that
render simple rules obsolete. Keeping this in
mind, the following types can be
distinguished:
Very little distortion is evident in all
symmetrical lenses in which the distribution
of refractive power to the front component
ahead of the aperture stop and the rear
component behind the aperture stop is very
well-balanced. The ideal case is represented
by special repro-lenses for a 1:1 scale which
are perfectly symmetrical such that the path
of rays is also symmetrical. Under these
circumstances there is simply no distortion.
Wide-angle lenses designed to be nearly
symmetrical, such as the Biogon,
approximate to this ideal as they have a short
backfocal distance that corresponds to their
short focal length. This allows for favorable
design conditions with specialized cameras.
The backfocal distance, i.e. the distance of
the last lens from the image plane, is a very
sensitive parameter. The older types of the
Biogon were even slightly better since they
did not need to take into consideration the
modern TTL exposure measurement and
were designed to have even shorter
backfocal distances than their modern
successors. But still, the TV distortion of the
new types is only approx. 0.2%, which is
negligible in normal imaging.
If we analyze the images by means of
metrology, mathematical corrections are
applied even to such small levels of
distortion. One goes so far as to no longer
assume the aberration resulting from
distortion to be rotationally symmetrical. The
distortion of our Biogon lenses for the NASA
was always measured in several azimuths.
Usually, short tele lenses, such as the
Planar 1.4/85, also show very little distortion,
especially when they are not really tele
designs at all, such as the Tele-Tessar 4/85
ZM.
Carl Zeiss
True telephoto lenses, in which positive
refractive powers predominate in the front
and
negative
refractive
powers
predominate in the back, tend to show
pincushion distortion.
In contrast, all retro-focus wide-angle
designs whose much extended backfocal
distance is needed for single-lens reflex
cameras, show barrel or moustache
distortion.
Due to their nature, the type of distortion
of zoom lenses varies with the focal
length since the distribution of refractive
powers changes in the process of
zooming. While barrel or moustache
distortion tends to predominate at the
short end of the range of focal lengths,
this tendency decreases towards longer
focal lengths or changes completely to
pincushion distortion. When the radial
distortion changes its sign in the process
of zooming, it must become very small in
certain favorable focal length ranges.
Within these ranges, the zoom lens is
often better than a comparable prime lens.
This may seem paradoxical to some of us
since it seems that one could simply
design a prime lens the same way as the
zoom lens and just dispense with the
movement of the zoom components.
However, this is undesirable since the
zoom lens is much larger than the
comparable prime lens.
This shows that compromises with regard
to distortion always need to take into
consideration the size and weight of a
lens. Often, a basically favorable lens
design has been rejected in order to stick
with the standard filter thread.
Lenses with a very large aperture are
usually associated with somewhat higher
distortion as compared to a more
moderate initial aperture if all other
parameters are identical. For this reason,
macro-photography lenses offer better
image geometry than high speed lenses
with the same focal length.
One should keep in mind that distortion is
a function of scale and can be changed by
any kind of optical attachments.
Camera Lens Division
18 / 28
TV-Distortion at long edge = 0.6 %
TV-Distortion at long edge = 1.1 %
12
16
20
0
0.0
-1.0
-0.1
-2.0
-0.2
-3.0
-0.3
-4.0
-0.4
4
8
12
16
20
0.0
0.0
-1.0
-0.1
-2.0
-0.2
-3.0
-0.3
-4.0
Off-axis distance [mm]
Slope [% / mm]-
8
Radial distortion [%]-
4
Slope [% / mm]-
Radial distortion [%]-
0
0.0
-0.4
Off-axis distance [mm]
Distortion of the fast standard lens Planar 1.4/50 ZF/ZE/ZK at taking distance infinity (top
and left graph) and at the minimum focus distance of 0.45 m
Carl Zeiss
Camera Lens Division
19 / 28
TV-Distortion at long edge = 0.3 %
TV-Distortion at long edge = 0.2 %
12
16
20
0
0.4
-1.0
0.2
-2.0
0.0
-3.0
-0.2
-4.0
-0.4
4
8
12
16
20
0.0
0.4
-1.0
0.2
-2.0
0.0
-3.0
-0.2
-4.0
Off-axis distance [mm]
Slope [% / mm]-
8
Radial distortion [%]-
4
Slope [% / mm]-
Radial distortion [%]-
0
0.0
-0.4
Off-axis distance [mm]
Comparison of Distagon 2.8/25 ZF (top and left graphs) and Biogon 2.8/25 ZM
(bottom and right): please note how small perceived distortion is even in the retrofocus lens.
But the Biogon is still better: the peak TV distortion of the Distagon is about 0.5% at 8 or
9mm distance from the centre, while the Biogon has a peak value of just over 0.2%.
Carl Zeiss
Camera Lens Division
20 / 28
Not all strange geometry is caused by distortion
Whether or not distortion bothers us in taking
photos depends mainly on the actual object
being photographed. In nature and landscape
photography straight lines are rare with the
result that we can tolerate larger errors since
we do not perceive them at all. This also
applies to portrait photography.
However, the more the world created by man
is shown in the image, the greater is the need
for distortion correction because man-made
objects are full of geometric lines. In
architecture and reproduction we tend to be
very critical of distortion. Even if the straight
lines are only background or of minor
significance, e.g. in photos of people in
rooms, next to furniture, etc., we find major
distortion to be bothersome.
The picture of this couple exhibits skewed
“eggheads” while the balloon in the middle
looks quite normal. The shot was taken
with a 16mm-lens on 35mm format.
No, rather it is a mistake made by the
photographer who took the picture setting
the focal length too short. It is an
inevitable feature of lenses based on
gnomonic, rectilinear projection that equal
angle intervals need to be imaged onto a
larger image distance towards the
periphery. This is demonstrated in the
following diagram using the example of a
15mm-lens for the 35 mm format:
On the other hand, a lens that is completely
free of all distortion does not solve all
problems of photography with wide-angle
lenses. For example in photos of groups of
people, those poor souls standing close to
the periphery of the picture appear weirdly
warped and skewed and are bedecked with
an ‘egg-shaped’ head. Is this due to yet
another lens aberration?
f= 15 mm
Shift= 0 mm
Field angle of view [°]
90
70
50
30
10
-24
-20
-16
-12
-8
-10
-4
0
-30
4
8
12
-50
-70
-90
16
20
3
2.8
2.6
2.4
2.2
2
24 1.8
1.6
1.4
1.2
1
du'/dW relative,
Elongation
FOV for image circle 43.2 mm: 110 degrees
Off-axis distance [mm]
Field angle
mm /° relative to mm /° min
spheres in a plane
Plot of the image equation for a rectilinear 15mm-lens: the black curve indicates the field
angle as a function of off-axis distance; this curve gets flatter towards the corners, which
indicates to us that distortion-free lenses can never be used to obtain a diagonal field
angle of 180°. The bold red curve is related to the slope of the black curve; its values are
obtained by calculating the reciprocal value of the slope of the black curve and relating it
to its value in the middle of the image. This means that the red curve tells us the factor by
which the image of the same field angle interval is larger at the edge as compared to the
middle: in the 15mm-lens, spherical objects at constant distance from the camera are
elongated by a factor of 3 in the radial direction, if they are imaged in the corner. This is
what causes the people to have "egg-shaped heads" as mentioned above. The thin red
line is for spheres in a plane which don’t have constant distance from the camera.
Carl Zeiss
Camera Lens Division
21 / 28
This effect does not occur in a flat two-dimensional object (left image) containing stripes of
equal width or circles because the lens sees a stripe at the periphery at a smaller angle than in
the middle. This decrease in angle fits exactly in the image equation, meaning that the
distortion-free lens reproduces the geometry of the flat object in a true- to-nature manner.
However, when dealing with 3-dimensional objects, e.g. spheres or heads, the situation is
different. The spheres are seen at the same angle regardless of where they are and therefore
the increase of the angle scale (red curve on previous page) in fact generates the clear radial
stretching of the image.
If one wishes to avoid this effect, one either
needs to use a longer focal length or, if this is
not possible because of cramped conditions,
to use a fisheye lens. Then however, once
again one has to accept the strong curvature
of straight lines. To optimize both at the same
time is ultimately impossible since there is no
way to image three-dimensional space onto
two-dimensional image in a true-to-nature
fashion - something has to give here.
These projection problems are some of the
challenges of creating the image when using
wide-angle lenses. Depending on the motif
chosen and individual taste they might render
the images either interesting or totally
unacceptable.
Carl Zeiss
The image of a brick wall looks very
strange when taken with a fisheye lens.
Camera Lens Division
22 / 28
We perceive convergent parallel lines as a
mistake mainly when they extend in the
vertical direction. This creates a feeling that a
building is toppling over. In contrast, very
strongly convergent lines in the vertical
direction suggest height to us, while
convergent parallel lines in the horizontal
direction are perceived as perfectly normal
since they match our visual experience.
The geometry of some motifs nicely matches
the imaging character of a fisheye lens and
the strong curvilinear distortion of lines at the
periphery is of minor importance.
Aside from these problems which are
particularly evident with round objects, there
also is the well-known phenomenon of
"convergent lines", i.e. lines converging when
the image plane and the object plane are not
parallel to each other. This is not a lens
aberration either.
To frame the complete building the camera
had to be tilted upwards. Converging vertical
lines are then a natural consequence and no
lens aberration.
Carl Zeiss
Nobody would assume here that the building
gets lower towards the right. We are used to
the experience that parallel lines of the
building converge to a common distant
vanishing point.
In this dynamic image composition the strong
vertical and horizontal convergence makes
some amount of distortion invisible.
Camera Lens Division
23 / 28
f= 24 mm
Shift= 0 mm
Field angle of view [°]
90
70
50
30
10
-24
-20
-16
-12
-8
-10
-4
0
-30
4
8
12
16
20
-50
-70
-90
3
2.8
2.6
2.4
2.2
2
24 1.8
1.6
1.4
1.2
1
du'/dW relative,
Elongation
FOV for image circle 43.2 mm: 84 degrees
Off-axis distance [mm]
Field angle
mm /° relative to mm /° min
spheres in a plane
Imaging function and angle scale factor for a 24mm-lens. The radial elongation is already
much less pronounced than with the 15mm focal length.
f= 24 mm
Shift= 10 mm
Field angle of view [°]
90
70
50
30
10
-24
-20
-16
-12
-8
-10
-4
0
-30
4
8
12
16
20
-50
-70
-90
3
2.8
2.6
2.4
2.2
2
24 1.8
1.6
1.4
1.2
1
du'/dW relative,
Elongation
FOV for image circle 43.2 mm: 79 degrees
Off-axis distance [mm]
Field angle
mm /° relative to mm /° min
spheres in a plane
If the 24mm-lens has a larger image circle diameter and its optical axis can be shifted
laterally by 10 mm, the angular scale factor in the lateral direction becomes
unsymmetrical and we have approx. the same situation as with the 15 mm lens on one
side, while on the opposite side the elongation is as low as with a 35mm-lens (below).
f= 35 mm
Shift= 0 mm
Field angle of view [°]
90
70
50
30
10
-24
-20
-16
-12
-8
-10
-4
0
-30
4
8
12
-50
-70
-90
16
20
3
2.8
2.6
2.4
2.2
2
24 1.8
1.6
1.4
1.2
1
du'/dW relative,
Elongation
FOV for image circle 43.2 mm: 63 degrees
Off-axis distance [mm]
Field angle
Carl Zeiss
mm /° relative to mm /° min
Camera Lens Division
spheres in a plane
24 / 28
Equi-solid-angle projection f= 16 mm
Shift= 0 mm
-24
-20
-16
-12
90
2.8
70
2.6
50
2.4
30
2.2
10
2
-10
-4
0
-30
-8
4
8
12
16
20
24 1.8
1.6
-50
1.4
-70
1.2
-90
1
du'/dW relative,
Elongation
Field angle of view [°]
FOV for image circle 43.2 mm: 170 degrees
Off-axis distance [mm]
Imaging function and angular scale factor for a format-filling fisheye lens with a so-called
“equi-solid-angle” projection: here, the angular scale factor decreases towards the
periphery to the same degree in which it increases in a rectilinear 35mm-lens. Thus the
distortion of spheres by the projection is minor - but the radial distortion value in the
corner of the image is approx. - 90 %. All lines that do not run through the middle of the
image exhibit a very strong barrel distortion.
Equidistant projection f= 8 mm
Shift= 0 mm
-16
-12
-8
-4
90
2.8
70
2.6
50
2.4
30
2.2
10
2
-10
0
4
8
12
16 1.8
-30
1.6
-50
1.4
-70
1.2
-90
1
du'/dW relative,
Elongation
Field angle of view [°]
FOV for image circle 24 mm: 172 degrees
Off-axis distance [mm]
Imaging function and angle scale factor for a fisheye lens producing a non-format-filling
circular image. This type of projection is called "equidistant”, since the image path of an
angle interval is equal throughout the image. Lenses of this type were developed mainly
for scientific applications for the measurement of angles in the object space by measuring
the distances in the image.
Carl Zeiss
Camera Lens Division
25 / 28
Two shots taken from the same point of view with a 24mm-shiftlens. For the image on the left
hand side the lens had been shifted to the left, while it was shifted to the right for the shot on the
right hand. To achieve about the same field of view in both cases the camera swivelled
accordingly, while the tripod position was identical.
Thus on the left the angle between the image plane and the longitudinal direction of the
locomotive is larger. This causes stronger convergence of horizontal lines. At the same time the
front wheel is imaged in the range of larger field angles of the lens, so that more radial elongation
applies to this side of the object. That is the reason for different shapes of the wheels: while the
rear ones are seen as a vertical ellipse due to the oblique view, the radial elongation distorts the
front wheel into a horizontally extended ellipse. This is really a matter of taste…
Another pair of shots with the 24mm-shiftlens from the same point of view. On the left hand side
the lens is in the normal centre position and the camera is parallel to the front surface of the
engine. On the right hand side the lens has been shifted downwards and the camera was tilted
upwards to see again about the same field of view. Three effects are evident: 1) the total object
field size has decreased a bit, 2) due to the oblique viewing direction and due to the increasing
radial elongation towards the lower frame edge the wheels are distorted into ellipses and their
axes are no longer in the centre, 3) the white bands on the kettle and the axes of the elliptic
shapes of the left and the right wheel exhibit convergence. Nobody would believe that this engine
could roll …
Carl Zeiss
Camera Lens Division
26 / 28
Correction by Software
We have now learned from several examples
that lens distortion cannot be perfectly
described by a single number. Instead we
have to understand a curve or a kind of map.
This sounds like plenty of numbers.
Fortunately things are not that bad, since the
distortion of many lenses can be described by
just two numbers with a sufficient accuracy.
All distortion functions which we have seen
above follow a quite simple equation:
a u '2 b u '4
In the simplest cases we have b=0 and then
the radial distortion D is proportional to the
square of the off-axis distance. If b is clearly
different from zero and a and b have different
sign then we have the case of a wave-type
distortion. The absolute error of the distorted
image point position is given by:
(known as the 5th-order approximation)
r
Sometimes there exist distortion functions
which don’t follow above simple equations
and then many correction tools fail:
0.5
a u '3 b u '5
The two constant parameters a and b in this
equation tell us everything about the
distortion of this lens – at a particular taking
distance. When distance and scale factor are
changed a and b change as well.
When the radial distortion curve of a lens is
known, e.g. from the maker’s data sheet, then
a and b can be calculated from just two points
on the curve.
The parameters can also be calculated from
image data: three points on an image line
close to the frame edge are basically enough.
In practice one uses more points, to account
for small errors of co-ordinate reading and
camera orientation. And of course one has to
assume that the line in the target is absolutely
straight.
If one knows the numbers a and b, one
knows the distortion error in each position of
the frame. And then this error can be
cancelled when the image is available in
terms of digital data.
Carl Zeiss
In addition the correction based on the
simple radial function assumes that the
distortion error is rotationally symmetric.
Real lenses have however also tangential
distortion, and the radial distortion is not
constant on a circle in the image. But
these effects are small and only of
importance
for
highly
accurate
measurement procedures. They can be
neglected in normal photography.
0
Radial distortion [%]-
D
Everybody is familiar with the change of
the digital image size, for instance to
match the file to a certain print size.
Interpolation algorithms are used for that
purpose. Distortion correction works in a
similar way, the size changes are just
variable within the frame.
-0.5
-1
-1.5
-2
-2.5
-3
0
0.2
0.4
0.6
0.8
Relative off-axis distance u'/u'max
Distortion of two lenses for mobile phone
cameras; the black curves are for two
different taking distances. Lenses with
strongly aspheric surfaces exhibit complex
distortion, which cannot be expressed as a
5th-order approximation.
Simple correction programs can only
handle the case b=0, pure barrel or
pincushion type of distortion. They cannot
correct moustache distortion perfectly.
Camera Lens Division
27 / 28
1
Converging lines and curvilinear distortion in
digital image files can be treated by suitable
software modules. Annoying errors of this
type can be cancelled semi-automatically or
manually. Some RAW conversion programs
(e.g. Canon Digital Photo Professional, Nikon
Capture NX2) offer an automatic correction of
distortion based on the EXIF data of the lens.
Independent
software
for
image
improvement (e.g. DXO Tools, Acolens,
PTLens) allows an individual definition of
correction profiles for various lenses.
Photoshop and Photoshop Elements
offer simple but quite effective correction
of normal barrel or pincushion distortion
with help of control sliders.
We plan to deal with these correction
procedures in more detail in one of the
future issues of CLN.
Manual distortion correction in Photoshop
Carl Zeiss
Camera Lens Division
28 / 28
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising