late
shift
late
tive,
hich
me,
cale
hor-
letal
the
nent
der.
y on
ler).
ssed
itact
e of
con-
1cial
veen
mm.
bled
too
vent
ling,
late
> by
on
ien-
d to
the
ints
phic
ing:
mm
mal
22)
Ops:
the
> on
The
vere
ob-
D at
aout
ture
The
tion
and
resolution dropped to the level of 75% at the centre,
and to 90% at the edges of the original camera
resolution. For the wide angle Flectogon 4/50 at the
relative aperture 1:4 the image resolution without
the reseau-glass was 37 lines at the centre of the
field of view and 10.5 I/mm at the edges. The
glass-plate caused the resolution drop by 8.7 l/mm
at the centre and by 0.6 l/mm at the edges. Relatively
resolution dropped to the level of 76% at the centre
and 94% at the edges comparing with the original
camera resolution (without the 1.6mm reseau glass
plate).
4. GEOMETRY OF PICTURES TAKEN WITH
THE ADAPTED RESEAU CAMERA
4. The reseau transformation accuracy
The first, very important step of processing of
photographs takea with a reseau camera, it is
elimination of systematic errors appearing during
image recording on the film. These systematic errors
are due to the film differential shrinkage, film non-
flattness in the moment of exposure, and by recording
of the image slightly out of the plane of fiducial
frame. The elimination, or at least reduction of
those systematic errors can be achieved by trans-
formation of photo-coordinates onto the reseau fidu-
cial frame. The transformation coefficients can be
calculated by fitting the coordinates of reseau-crosses
surveyed on the photograph onto the coordinates
of crosses of reseau master pattern surveyed directly
on the reseau plate (the reseau master pattern was
surveyed with the accuracy +0.8um using Zeiss Jena
Ascorecord).
In the presented experiment we have examined 12
photographs taken with our 6X6 reseau camera, to
get answer to the following questions:
e by what value of deformation are influenced the
photographs,
e which type of transformations would be the best
for photo-coordinates refinement,
e how the number and distribution of the reseau
active crosses influences the accuracy of refine-
ment procedure (as an active reseau cross is
meant such a czoss of the reseau pattern which
is actually used as a control point in the coor-
dinate refinement procedure).
To get answer to the above questions the experimental
12 photographs selected from three films (ORWO NP
20) have been measured. On each photograph there
should be registered 49 reseau crosses, but on a few
pictures one cross could be omitted if imaged on the
black image fragment and badly visible, and such
picture could be partially excluded from the analysis.
The reseau crosses were surveyed with the accuracy
+2um using the Zeiss - Jena Stecometer. The infor-
mation about the size of errors caused by the above
described deformations can be achieved by isometric
transformation of coordinates of all the 49 crosses of
the photograph onto the coordinates of all the 49
crosses of the reseau master pattern. The residual
deviations on all this active points will describe optimally
the quality of the identity of the both compared sets
of points, as the isometric transformation does not
change size or shape of the transformed picture. The
results of this isometric transformations shows the
average standard error calculated for the 12 experimen-
tal frames is +36.3um. The analysis of errors shows
that the residual deformations fluctuate up to 50% of
average standard deviation even for pictures recorded
within one piece of film. There was found experimentally
that the best description of deformations give: bilinear
transformation and projective transformation. Compar-
ing the results of those two transformations when
applied to our test photographs it was noticed that
the average standard deviations are in both cases
identical avr. mx=*24um, avr. my=*27um, avr.
mp = +3.7um. However, the bilinear transformation is
less sensitive to noises and can be used to describe
the photograph deformations. To check the property
of the deformation there were calculated separately
the bilinear transformations for the points distributed
over the area of each quarter of the frame. The
calculations ranged over only those quarters of each
of the 12 experimental photographs on which there
was possible to measure all the 16 reseau crosses (all
the 16 crosses were active in the transformation). The
calculation results are shown in table 1. And below
are the average values of standard deviations, and their
errors, for the separate quarters:
— for the I quarter avr.mp=*19um, o==0.4um,
— for the II quarter avr.mp — €£2.3um, g— x 0.6um,
— for the III quarter avr.mp = t1.8um, g — x0.3um,
— for the IV quarter avr.mp ^ £3.1um, g — x0.4um.
The results show that the average value of standard
deviations taken from the separate transformation of each
quarter is smaller than the one for the full frame trans-
formed at once. The best result was achieved for the
I-th quarter, and the poorest for the IV-th quarter.
For average case of practical applications of the reseau
camera it would be rather costly to measure all the
49 crosses on each reseau frame. So the experimental
bilinear transformation was made for all the 49 crosses,
using exclusively 9 regularly distributed active crosses,
but only 4 of them at a time. When transforming
separately each quarter area of the photograph, we
got different results, which points out on differences
of the physical model of errors on different portions
of the photograph.