ADA091 238 DEFENCE RESEARCH ESTABLISHMENT VALCARTIER (QUEBEC) FIG
14/5 IMAGE*ENHANCEMENT BY SPECTRAL RATIOING,(U) JUN 80 J F
BOULTER
UNCLASSIFIED DREV-R-4170/80 NL*E ' IIIIIIIIFI
-- END
UNLASSIFIED UNLIMITED DISTRIBUTION .
CRDV RAPPORT 4170/80 DREV REPORT 4170/80 DOSSIER: 3621J-007 FILE:
3621J-007 JUIN 1980 JUNE 1980
AD A0128 LEVELf0
J.F. Boulter
BUREAU - RECHERCHE El DEVELOPPEMENT RESEARCH AND DEVELOPMENT BRANCH
MINISTERE OE LA DEFENSE NATIONALE DEPARTMENT OF NATIONAL
DEFENCE
CANADA NON CLASSIFIE CANADA
/ IMAGE ENHANCEMENT BY PECTRAL RATIOING
by
(/0
DEFENCE RESEARCH ESTABLISHMENT
RESUME
Dans une image produite par la rdflexion de la radiation, le
contraste r6sultant des diff~rents types de surface est souvent
plus important que celui caus6 par des diff6rences de
l'illumination ou de l'orientation des surfaces. Ce contraste peut
6tre difficile 5 pr6voir, en pratique. On d6montre que, dans
plusieurs cas, le contraste produit par les differences de
l'illumination ou de l'orientation peut atre r~duit ou 61imin6 si
on divise une image obtenue dans une bande spectrale par celle
provenant d'une bande spectrale diffdrente. On applique cette
technique pour rehausser des images des v~hicules militaires
enregistr~es sur un film en couleur de m~me que celles d'objets
faits par l'homme par rapport h la v~g6tation sur une image obtenue
avec balayeur multispectral. (NC)
ABSTRACT
In a reflectivity image, contrast produced by differences in
surface material is often of more interest than that produced by
differences in surface orientation or illumination. The latter
factors may be difficult to predict or to allow for in practice. We
show that, in certain cases, contrast caused by differences in
surface orientation and illumination may be reduced or eliminated
by dividing an image obtained in one spectral band by one obtained
in a different spectral band. We illustrate this technique by
enhancing the images of military vehicles recorded on color film,
and by enhancing man-made objects in an aerial multispectral
scanner image relative to vegetation. (U)
jccti'.ton Po r
taxmounc cd
5.0 REFERENJCES
...................................................... 15
1.0 INTROJDUCTION4
Contrast in an image formed with radiation reflected from an
area
containing a target of interest depends on the orientation,
reflectivity
and illumnination of the surfaces in the target and background
regions.
In some cases, contrast due to differences in surface material is
more
important than that due to differences in orientation or
illumination.
The latter factors may be difficult to predict or to allow for
in
practice, and the contrast that they produce may have undesired
effects.
Glint or shadowed regions can reduce the performance of an
automatic
target acquisition or tracking system, for example. In this report
we
describe a method, which can be applied in certain cases, for
enhancing
the contrast due to differences in surface material and for
reducing
that due to the orientation or illumination of the reflecting
surfaces.
The method requires 2 registered images of the target area
obtained over different spectral ranges. Under certain assumptions,
the
ratio of the 2 images is independent of the orientation and
the
illumination of the surfaces, and depends only on their
spectral
reflectivities. Ideally, all contrast in the ratio image will be
due
only to differences in surface material.
In Sect. 2.0, we describe the theory of the processing and
briefly discuss how the spectral responses of the 2 detectors
should be
chosen to maintain or to improve the overall contrast of the
target
against its background while the orientation and illumination
effects
are reduced. In Sect. 3.0, we use the technique to enhance the
images
of targets consisting of military vehicles against backgrounds
of
UNCLASSIFIED 2
foliage and ground terrain, and to enhance man-made structures in
aerial
imagery relative to vegetation. Some possible applications of
the
technique for improving images intended for subsequent human or
machine
analysis are given in Sect. 4.0.
This work was performed at DREV during the period
February-May
1979 under PCN 21J03, Imaging Seekers and PCN 21J07, Target
Acquisition.
2.0 THEORY
Consider a ray of illumination reflected from a target into
the
field of view of a point detector. The direction of the incident
ray is
specified by the 2-component vector i and that of the reflected ray
by
the 2-component vector . The shape of the target is described by
the
normal to its surface N( ), and its surface reflectivity by the
function
R( X ,e,-). The 2-component vectors i and represent the angles
between
the normal to the target surface and the directions of the incident
and
the reflected rays. Figure 1 illustrates the projection of
this
geometry onto a plane that passes through the point of reflection
on the
target.
We assume that the illumination function I(X,\) is the same
for
all points on the target and that it depends only on the wavelength
and
on the direction of incidence. Consider a small area on the target
that
corresponds to the incremental solid angle do that is within the
field
of view of a linear detector. This area receives illumination
within a
solid angle dr and a spectral range dX . The corresponding
detector
ovtput dW is proportional to the product of the illuminance, the
surface
reflectivity and the spectral response of the detector D(X):
UNCLASSIFIED 3
UNCLASSIFIED 4
dW(,X, ,a) = I(x,) R(, ,j) D(X) do dO dX [1]
We make 2 assumptions that will be more or less true in
practice.
The first one is that the reflectivity R can be written as the
product
of a part that depends only on wavelength, RI ( X), and a part
that
depends only on the angles of incidence and reflectance, R2 (0,4$).
The
second one is that the illumination I can be written as the product
of a
constant 10, a part that depends only on the wavelength, 11 (X),
and a
part that depends only on the illumination direction, I2 (P).
2
We obtain the total detector output W by integrating eq. 1
over
all illumination directions A, over the detector field of view B
and
over all wavelengths . The assumed separability of the illumination
and
reflectivity functions permits the wavelength and spatial
integrations
to be performed independently of one another:
W = oi,(X) R1 (X) D(X) dX ]f 2(E) R2( , ) da d8 [2]
0 A B
Suppose 2 coincident detectors with different spectral
responses
Di and D. form registered images of the same target area. At all
points
in the 2 images, the ratio of the detector responses: 00
W A f1(X) R(A) D i(A) dX0 [3]
(XC) RI(CN).Dj (X) dx
depends only on the integrals over wavelength - the spatial
integrals
are the same for both detectors and cancel in the ratio. The ratio
is
LNCIASSIF'IED 5
independent of the absolute illumination level IOF the
angular
distribution of the illumination 12 the angular reflectivity
function
of the surface R 2. and of the orientation or irregularities in
the
surface. F'or a given pair of detectors D. and D.j, and for
illuminance
with spectral content IVthe ratio depends only on the
spectral
reflectivity of the surface R. Contrast in the ratio image
depends
only on differences in the spectral reflectivies of the surfaces,
and
not on their orientations or illumninances.
The spectral responses of the 2 detectors must be chosen so as
to
ensure that the target of interest has sufficient contrast against
its
background in the ratio image. It is possible for the ratio given
by
eq. 3 to have the same value for a background region as for a
target
region with an entirely different surface; in such a case, the
target
has no contrast against its background. The choices for the
detector
responses Di and D . that optimize the tradeoff between
target-background
contrast and signal-to-noise ratio must take account of the noise
levels
present on each detector output as a function of wavelength as well
as
the distributions of 1I1 and R1 . In performing this optimization,
the
overall detector spectral responses used in eq. 3 can be
synthesized by
combining the outputs of several detectors with different
spectral
responses.
The physical model represented by eq. 1, and the assumed
separability of the illumination and reflection processes
into
wavelength dependent and wavelength independent parts, is
only
approximately valid in practice. For example, if the target
is
illuminated by fnuitiple sources with different spectral contents
(e.g.
UNCLASSIFIED 6
directly by the sun as well as indirectly by the light reflected
from a
nearby "colored" surface), then the illumination cannot be
expressed as
the product of independent parts. Also, particularly for
highly
specular surfaces, the angular reflectivity R2 may depend
significantly
on wavelength, or the spectral reflectivity RI , on the angles
of
incidence and reflectance (Ref. I). Furthermore, if either detector
is
not linear, the ratio may not be independent of surface orientation
and
illumination. However, as shown in Sect. 3.0, in some situations
of
practical interest, the assumptions are sufficiently true to
allow
useful target enhancement based on this principle.
3.0 TARGET ENHANCEMENT EXAMPLES
Color images of a helicopter and of the front and side views of
a
canvas-back truck were recorded on Kodachrome-64 color
transparency
film. Regions of the images were digitized to 256- by
256-element
resolution with a high quality vidicon camera interfaced to
an
interactive digital image-processing system (Ref. 2). This vidicon
is
linear within 2% over a 3-decade dynamic range. Each image
was
digitized 4 times: first without a color filter and then with 3
Kodak
color filters (types 25 red, 58 green and 47 blue) inserted
individually
between the camera and the transparency. Figures 2(a), (c) and (e)
show
the original digitized images obtained without a color filter
whereas
Figs. 2(b), (d) and (f) show the ratios of 2 images obtained
with
different color filters.
In general, optimum choice of the detector spectral responses
rrquires a priori knowledge of the spectral content of the
radiation
UNCILASSIFIED 7
FIGURE 2 - Three original images are given in (a), (c) and (e).
Calculating the ratio of the images obtained in two different
spectral bands reduces contrast due to differences in surface
illumination and orientation, and enhances contrast due to
differences in surface material, as shown in (b), (d) and
(fL.
UNCLASSIFIED 8
reflected from the target and the background regions. As shown in
Ref.
3, the reflectivity of "dark green paint" varies between 4% and 12%
over
the spectral range 0.4 to 1.8 pm. The reflectivity of green
vegetation
lies within the same limits over the range 0.4 to 0.65 Am (violet
to
scarlet), but rises rapidly to approximately 65% over the range 0.7
to
1.2 pm (red to near infrared).
We used blue-to-red and blue-to-green ratios to ensure that
the
targets had a higher gray level than their backgrounds in the
ratio
images. The 2 ratios produced similar target enhancements but
different
noise levels due to film granularity. The blue-to-red ratio gave
a
slightly lower noise level for the truck images, but the
blue-to-green
ratio was preferable for the helicopter one. We tried using
various
weighted sums and differences of the 3 color components to
simulate
different detector spectral responses. Interactive adjustment of
these
combinations allowed us to obtain better target enhancements than
those
shown in Fig. 2, but we made no analytic attempt to optimize
the
results.
The 3 ratio images given in Fig. 2 show a clear reduction in
the
effects of illumination and target surface orientation in
comparison
with the original images. Highlights on the rear wing and on the
top of
the helicopter, for example, appear with almost the same gray level
as
the shadowed back and side regions. Similarly, in the front and
side
views of the truck, the strongly reflecting area on the engine hood
has
about the same gray level as the darker front and side regions in
the
ratio images. This results because the metallic surfaces of the
targets
are coated with the same paint and have the same spectral
UNCIASSIFIED 9
reflectivities. The spectral reflectivity of the canvas is
also
relatively uniform over its surface, and the contrast due to folds
is
significantly reduced in the ratio image.
Calculating the ratio did not remove all of the contrast
apparently due to surface orientation and illumnination effects. In
the
truck images, for example, the highlight on the top of the
canvas
covering remains. In most cases, such occurrences correspond
to
overexposed or underexposed regions on the film where the
relationship
between exposure and film density may be different for the 3 dye
layers.
In other cases, the remaining contrast may be due to true
surface
differences or to discoloration caused by weathering or by a
coating of
dust or dirt.
The texture of the canvas covering of the truck is different
from
that of the painted metallic surfaces, and this produces a
large
difference in gray level in the original images. It is interesting
to
note that because the "color" of the 2 materials is approximately
the
same olive drab, both appear with almost the same gray level in
the
ratio images. The effect of surface texture is contained in the
spatial
integrals in eq. 2, and these cancel in the ratio.
other man-made objects, present in the images shown in Fig.
2,
are also enhanced relative to the natural backgrounds when the
ratio of
the shorter wavelength image to the longer wavelength image is
taken.
In the side view of the truck, for example, a second set of power
cables
in the upper third of the image is invisible in the original,
but
contrauts easily with the background in the ratio. Similarly,
an
UNCLASSIFIED 10
aluminum-painted fence in the upper part of the same image is
strongly
enhanced in the ratio.
We summed 4 bands from an ERIM M7 multispectral scanner over
the
visible range 0.40 to 0.72 urn to produce the image shown in Fig.
3(a).
Figure 3(b) gives the ratio of the image obtained in the 0.40 to
0.44 urm
band to that obtained in the 0.66 to 0.84 Arn band. Close
inspection of
the ratio image shows that most of the man-made structures have
been
strongly enhanced relative to the vegetation. All buildings
and
roadways have a high gray level that contrasts easily with the
darker
vegetation. For example, a nearly vertical roadway located just to
the
left of center is barely visible in the original image, but is
clearly
defined in the ratio image. Similarly, houses and roadways in
the
residential area in the lower right quarter of the image are
strongly
enhanced. Contrast caused by differences in vegetation or by
shadows
(the sun angle is low) is also significantly reduced.
The man-made objects are enhanced in the ratio because the
vegetation typically reflects 3 times more strongly than the
man-made
structures in the range 0.66 to 0.84 Am', whereas their
reflectivities
are approximately equal in the range 0.40 to 0.44 urn. The
numerous
bright points in both the original and ratio images are caused by
lights
or other active sources and do not result from reflection of
radiation.
The extent to which this approach may be useful as a general method
for
enhancing man-made targets against natural backgrounds in
visible-light
imagery has yet to be determined.
UNC[ASSIF'IED
> 0-
4.0 DISCUSSION AND CONCLUSION
The objective of the present enhancement is to reduce contrast due
to differences in surface illumination and orientation, since
these
effects are often unpredictable in practice, and to enhance
contrast due
to differences in surface material, the contrast of man-made
objects
against natural backgrounds, for example. The processing
removes
information, but the assumption is that the resulting image is
rendered
more intelligible by the attenuation of the unpredictable
components.
Humans are experienced in using information obtained from
changes
in illumination or surface orientation, so the enhancement may be
of
most interest for reducing the complexity of images intended for
less
sophisticated machine analysis. The performance of some types
of
automatic target acquisition or tracking systems, for example, may
be
improved by applying such enhancement.
A non-updating correlation tracking system attempts to locate
a
previously stored reference image of a target in the input scene
by
correlation (Refs. 4 and 5). Good performance depends on the
target
illumination and orientation being the same in the reference and
input
scenes, but this may be difficult to guarantee in practice. If
ratio
images are used for the reference and input scenes, the correlation
may
be made less sensitive to such effects.
One type of automatic target acquisition system segments a
target
from its background based on the values of local features such as
gray
level (Refs. 6-9) or texture (Refs. 10-12). Properties of the
segmented
UNICLAIFIED
regions, such as connectivity, area, shape, context etc., are then
used
to attempt to locate or classify the targets of interest (Ref. 13).
In
the original target images shown in Fig. 2, the targets cannot
be
segmented from the background regions by setting a single
gray-level
threshold level. Some areas of the targets are darker than
the
background, whereas others are lighter than it. In the ratio
images,
however, most areas of the targets are clearly higher than
the
background region in gray level, and segmentation using the
color-ratio
feature would more closely preserve the shape of the target. We
plan to
do further work to evaluate, by simulation, the possible uses of
this
technique with specific target acquisition and tracking
algorithms.
Other target analysis operations, such as those based on
extracting a characteristic target signature (e.g. as obtained with
the
2-dimensional power spectrum, Ref. 14), may give better results if
the
unpredictable illumination and orientation effects are first
removed
from the target image. There may also be applications in the
general
areas of pattern recognition and image understanding (Ref. 15).
In
particular, the information obtained from the ratio image could
be
useful for detecting homogeneous regions or objects that differ in
gray
level because of differences in surface illumination or
orientation.
Enhancement of image displays intended for human
interpretation
is another possible area of military interest. The various
bands
available from a multispectral imaging system, for example, could
be
suitably combined to synthesize the 2 detector responses required
to
enhance the contrast of a particular type of target against a
specific
type of background. However, the images obtained in all spectral
bands
UNCLASSIFIED 14
must be formed by reflection of radiation. An image formed by
the
emission of infrared radiation in the 9-13 Am band, for example,
could
not be used.
Edition, Longman Group Limited, London, 1970.
2. Boulter, J.F., "Interactive Digital Image Restoration and
Enhancement", Comp. Graphics and Image Proc., Vol. 9, pp.
301-312, December 1979. Also published as DREV R-4143/79,
June
1979, UNCLASSIFIED.
Electronics and Electron Physics, Vol. 28B, Academic Press,
New
York, 1969.
Similarity Detection", DREV R-4097/77, November 1977,
UNCLASSIFIED.
Compensation for Magnification and Rotation", DREV R-4140/79,
July 1979, UNCLASSIFIED.
6. S~vigny, L., "Le traitement des images et Vacquisition
d'objectif dans 1'infrarouge", DREV R-4050/76, avril 1976,
NON
CIASSIFIE.
d'objectif infrarouge dans un contexte sol-sol", DREV
R-4081/77,
juin 1977, NON CIASSIFIE.
d'objectif en infrarouge: Nouvel algorithme de detection",
DREV
R-4099/78, mars 1979, NON CEASSIFIE.
9. Panda, D.P. and Rosenfeld, A., "Image Segmentation by
Pixel
Classification in (Gray Level, Edge Value) Space", IEEE Trans.
on
Computers, Vol. C-27, pp. 875-879, September 1978.
10. Mitchell, O.R., Myers, C.R. and Boyne, W., "A Max-Min Measure
for
Image Texture Analysis", IEEE Trans. on Computers, pp.
408-414,
April 1977.
in FLIR Imagery", Proc. IEEE Computer Society Conference on
Pattern Recognition and Image Processing (Chicago, May 31 -
June
2, 1978), IEEE, New York, NY, 1978.
12. Boulter, J.F., "Automatic Target Detection Using Textural
Information", DREV R-4152/79, August 1979, UNCLASSIFIED.
13. S~vigny, L., "Extracteurs s~quentiels pour lacquisition
de
cibles sur images", DREV R-4153/79, aout 1979, NON CLASSIFIE.
ia
1975, UNCLASSIFIED.
15. Pratt, W.K., "Digital Image Processing", John Wiley and Sons,
New
York, 1978.
0* 0 - W '.n 4) D( C En . CU
c 0 ) 4Q)I0 0 0)) 0 .- C Vr -) L)))0
C- (D c cu 0 c 0 0 - (D(L C 4) d
(f = 0)0)=4),)40 4) > ) ) 0 t 0
'0 0) ~ :) cc 4)'4 v) ~ 00 004
00 ' r) a~)~ 0' a '4) 4) cc 2 (ft~(u 0. 5-d)(.004 0 (. 0 0 5 0
'
C.)~ ~~~ ~ c 4)5). (n)504 C.W )-. ~ f50 .0 4) ID L. ) c) .0 -6 L)
-4.4).4 Mf C)
c cu (t OcE 0.4 g. o- 0 .f
.0 L)4) d) . 05 4 .CD U. 01 '' c 4) - 0.(4 )
cn - m 0 -j 4) (.0 d )
5- 04) U) 00) ) C' d 5- u.4~)wma) p '420)) V. d0-00.')E . W40 0.
00-'000
D c0 (0 )V. 4)0 UJ'. ) 4).j -')0> 02~~~ Df '(D C-. '4.C) 0) (t 0
) 0 d ' 4)
..( 0 4 (t. 0))0- 5- c c4 0 A d 4)0t. 04)) -a c Z i 0)5 0 V E a ca
0to 0'c u 4) " a m ca c c 11.
0 ~ ~ ~ 4 (ftQ .Cs.t'2 -4)) ) 0 CUf .4.-.4)40 4)) .C)0 4) 00 W0
)4'-.> 4U 0)0 4) 0L) 00 d).44
cu fn0. =40400 - .. 4000. 4w4)02)0
'0L4 V 4 m ',Wcc - m4) *04f '0 '4
a a.440 01 '0 S.- .4))
-~ ~ ~ 4 0(C) )..(t (10 )~f
* ~ ~ ~ ~ r -a. w4. 540 . 0 -'2* 5 4. .0 .0 -) 42
a 0 . 0 0' 4)'4 E 0cc . 0 -. )) ,0 0o00-
(f - (n0Q.. ca 00 (ft L 4) -0.' a)'0 E- SO '4 w~)0' 4 01)0 0-
o
w4C ) 4))5. ' (ft 0 . 4) c L)' C L w L)0. 4 )); .4 C. M. 4)0.
d))..4
4) 0' -) 0)w Ef4.)" a)''4 4) 4)0 2~~~~~ CC 4)...l) (fL(tZZ ) M
)..J) v) .)
0- ' )-40-1 0)) m- '0 .)) 4)04)-V r
LO > 4 (ft> cc a 4 4) c .0-'- 4, c)> 504 0 d) 4
.r w 5U - ..')4 0 - 4)= *l .0 5O > c m(4) . =)4.- W 0 r,0 0i E)
4 '05. 0.- -C (L 0 L) L). '05- 0.-.
L) c .cr00 CIO) 0 0. c)0 05. 0. E
") U .d) .m. F 0 ) -4) = 45 W) W U 4) .- 04) cr a) c 4
5. *4))0 w- 0 0r C.0) fn u0j=L ) m 0f .G.
- 0. 0) '04 U) d0. - ) Z 0. a, a04 ) U) 54)0
- ~ 9 ,...M C.4 4' 'm d ) m - 4 5 44, )
6I 01.4 5- E.04 04 L). w' d). 5.w 0 ) 0) u cc 0/ ) '0 )))50 4) '
.(D W'04-4)44))05 cua , )-
'424 V. d) 'Do m m M - m '42a) 0. .' 0 . 4) 0
-h O :3- 4))S)5.4- cc 4( CD 4) M0- 4)4 4)-
C. ) 4) co U) W- C. 4) cc a)4 4.7 0 CL0 4) Lz COW 5- C '4 M -(D W4
4) CL' gE
) C.0 - >-
Z~4 4)5 0' d) 00 4 M.5C.. 0 d)~ 4) C) 0 ))-U4 0 L cc M) r44. .m 04)
o-444 W ) *'.5(n-X4). _4)U .W
f to0 w) 04 V 4)45> Q m) 4) 0))4 4.-5
ID ;- 5- 40 w r- 1 ) 04 a) ccC. A5 c)4) w~0 04
0m 4) 4). CL 0 WV 0.~ a, CL 4 01. n( 5'. z).-
m -: 0.4. 0 E r-. '00 ... 910 "05 4) CC. '0.. 0- L 0 '4- )5 417 0 ~
~ ~ C 4) 0) .4 L.04. ) -00) 4-4o .. 44.
0i cm w r z )5 -. 4)V w m C- WC0r0CID) 4)'L'4) 0 4) 1') cc 3. 0l
.40 - 4)cc ..
co 1.4)0 d) ) c 4)c )d )d - )4 ol 4) -) 'm..=-C.W.. WOO40.J4 4)
C
; . 54C 44- :)4) 0 c 0 " 4)44-.I CD = ) 4D -.. '5 ..... 5- 4) C)
.)) 5 U w- 5- .'4.-4. )
c)~ c .,) >5 m. c CL -4) > :3 c-. 0 -..
(W 0 d) -0 L) Wc).4.CC D. m-0 U0 a 4 cC).C U6-4 .C0)d)0
0)0 a . C C.. 0)C .0. MW a- z. 4) 0.- 0. C- -CC I ... >.
4.- C-) CC C--C..C 04) - *- 4) ) C-DU ))) CC-o 2.c X .~ U I-3C IS
U) (
CDC E *0Cf.C S.C 0 a .. a)- C0 0.C *..C-CCC.l EC d.)CnCC-E
>.OC C- oC >. C - DC C)C- .. 0C.. 0).U C- -. C).-C Do i .. '
CJ a.Z 0 C U- C)0-D
. CC *-CCCzD .ccC *..C 0.L CDC. ) a o-. w0 C n* ) C, cc 0. fD~ .C a
-0) >, CO) D ~ 00 C d, >"-E) .
(D L, ) 4- a, Dl m UD~0 .0 C C-
CD UD)0- CD D4 C) Em M t..COC.. CD 0)D4 1=O 4) E- C- .. V.- .. aC-
ED a) lC u E". ... 4.. E w
m)f l C-. d) VEc 00 -0L C U.-. W0E m D a) .0 4-CC a)a,4 c m(L C cc
C-C - C , d
c r) 0 .- CC) D .z C C c D-.C0 cC
co * -CC 7m > CM 0c7.CD .C)> U0~ L_ C -00 E c 0C-UDC3)..v 'a
.C .0 C D 0
c a 0).C0D.C'C .cD 0)E-O) C .
.c 0 4 ) U CD -0C.0-C). ) CD c_( CD - .c , d ) z . 1C.CC D0 0) 0) -
C) >= " -EC'iCCC0 C.>0
m C. CD 2 0 00w0 VC-m -0 C4) CD MC C C- 1. )0 . 41 MO ).C >1.
cc0 =- 0) - )4.C 1c
-~~~ oC-. C. -... D C0C 0) - M-- M.. *.. C .0 DC 1. L) C)- eC-C) )
D C. ~0) C) >.CC- ,_- m )00 M.
C - m 0) c .0 0 " LC 0C c C) 0) .- 0 0.c c CL 0C
U) C ) d na) >1 c 0) >.-CC w) C3 4 / C.00.. ) >1
c-..
.. J~I' EL c' .C a)DC.C ~E) > 40 C.C- co a) L) C - C)Q>. C 0
)o Li ) L) U) C.- C) 0 0)
Z 0 d) 0.. m 1 CDO)C . 0 - C)CflC0 c DC )C. m
W~ 0 C- 0-. ... 0C.0 00 d C ~ 11C E)- .. CM.0 DC 0= U 0 C-D.CD CL
CC 0 j ) 'a) C DCCD (0C
D 0 C)-C U- 6-)..-. = z 0 V) C- c)C).0 C
> cc a, * 0C dD)C..-1C.0 0C4 1 >C co CD . (D 0)c.-*-*.. 0 . -
L 0 4 C _ fn)-CDC.- ) -n1 Ed 0 C CE)DU).~ C)W
"D Cm 1) -. C.. m m DD0 "I "a m C: . - ) CD CC0c - 1.-. CC g.0 a--
OC I. C3 c) to C U)) 6 - -. 0 .
m C- c0 ) 4)-0-C D DD CU 0) C)=Cw m 0Wt; ... Ca)
1. CL) D C-C.->) . CL ) U CL) 0)L 4 C-C > 0CL W C 0)o C7 cc d
f..-.DCC--0) cr co~ 4).> ~ .0 U m)2
0 ) to D.C WDE-CD ..- >0 0 0 40 E'C D-C -
0) - :, D OCD.
0)0)C C-w. enC0C .w t
C. .S C d4)C- C.C.. C D-w.tn
4)cQ ). ... C) C..0C .C 20'... MD- 0 C..
c~-0 U Cccd) 0) X.).. > 0: I) a) . 0
t0 .. C C- CC. w0 .0 c
C- .. C) 4-C (D-EU M 0...C C L)CD C- UCCO).C.0 E.
cu 4) d)0C-D . U0 CD >.C-.cicC CDCCD C C. C4 M C, U0 -0) .. 0)
0( C. .. ,a-C0) 0C.
4)C-C .C40 0)'0 -U- tD-C0.Cn
L)Cr 0' C. 0)D C) D r) . -
-. d) 0-.C 0)4) C 0 6. 000 0 0) w-.0C..4 -C C) 0-C.0) I., .. C-%_ 4
. C UrC > 00) 0 1 0) a'00CC.
CD) 0 .C 0.. .4 I- C< c6 0).-O) .C C> Li CD .00>C -It 13
e' fC C ) .to CZ
.00 3, "Z o- .00 Or C o .. , 00.0C
-~ ~~~~~~~ 0).-..C-40 >I). - 0C-..C - WE0 0 4) *.d O C-0 0) C.
d) wC 0 C
c t o -v c > .0 c 0.240 - ; U 0) 01. .4 )C 0) W.4 .a- 9) C)
>'C-)C 0
C) a) a6) . C C 0) (a4) W_) C9 jf C.-00)- C..6) C M) >C0) .. 0
.DCC0)0 0 ~C .- - ,L a, .6 4)0 CCC
a' U> .0 ow
o 9 m)U -ca 0 W0 U C1>. C - CD C CC. LI r) L, C-4 0 d)4 L) -.C .
.) 00>W )6 10 .60o. CC) a)~ I d, c -'CD 4)6L-4
0r c 6.. 0 > C-4.C) LnC)CD
0j M. C- UU..L .. -A. -C d) C- C)CCD.0 - 4) .6 DC.C C -1C-
go0, 0)-.4 0 ~ Q C x m) * 4 U).....0 z 4 .0V. L.) 0 to m 0) -. 0 C
C ). .cC
o )C.- a)- ) c-c& C- U 1 w .. 0 - C9)C w) C - 0 C .V 0 C- , 0.
a, - I ) )CCD.V 0 M) * 40) 0 - , g.C- m) C_ 0 m) . CDC-m
d.0 Q..4) C.0) W. ... 0 .. D -C-
->.~~~c -D4.t.C4 - C)> M C D .C C-0