вход по аккаунту


Image EnhancementЧChemical Digital Visual.

код для вставкиСкачать
Volume 28
Number 12
December 1989
Pages 1601- 1766
International Edition in English
Image Enhancement-Chemical,
Digital, Visual
By Joachim Lohmann*
Digital image processing has become well known through the high quality of some spectacular
satellite images. It has developed into a discipline in its own right within information science
and now finds application, not only in space travel, but also in such diverse fields as, for
example, robotics, medical imaging, and television technology. Image processing also occurs,
however, in photographic films and biological visual systems. In these systems, too, recorded
images are manipulated for better recognition. This poses the question as to whether similar
principles of image processing are operative in these different image handling systems.
1. Introduction
The term “digital image processing” refers to the computer-aided modification of images and encompasses image enhancement (modification for better visual recognizability),
image restoration (elimination of a systematic distortion),
and image analysis (extraction of information) (Fig. 1). Several monographs on digital image processing are available,[’ -51
Generally the photographic film is known only as a means
for image recording. But during chemical development the
primary image--the latent image--is not only made visible
but is improved in many respects. This forms the basis of
modern color-negative films, which combine excellent reproduction of detail with high color saturation. The model for
both digital image processing and, in many respects, for the
photographic film is the visual system. It performs all the
steps from image recording and image enhancement to image interpretation.
In this review the three systems-the chemical film, digital
image processing, and the visual system--are compared solely from the standpoint of image enhancement. Naturally,
Dr. J. Lohinann
Agfa-Gevaert AG
5090 Leverkusen I (FRG)
Image Capture
Image Processing
retinal image
Image Analysis
image analysis
image analysis
Image Enhancement
Fig. 1. Steps of image manipulation in the Photographic film. in digital image
processing and in the visual system.
very different physical and chemical mechanisms are operative in these systems, i.e. the image information is encoded in
different languages; consequently, the comparison relates to
the structure of these languages, to the algorithms of image
processing. The discussion is restricted to the improvement
of three basic image properties, namely noise (Section 3),
sharpness (Section 4) and color reproduction (Section 5).
2. Architecture of the Systems
contain dye-precursors, the couplers. Upon exposure to
light, clusters of silver atoms Ag,-the latent-image specksare formed (Fig. 3). During the subsequent photographic
development these silver clusters catalyze a redox reaction
between the silver halide and the developer. The oxidized
developer in turn reacts with the coupler to the image dye.
In the following the essential elements of the three systems
are briefly described.
2.1. Color Negative Film
Figure 2 shows the common three-layer structure of a
color negative film. Each layer contains silver halide crystals
which are sensitive to blue light or which are sensitized by
appropriate dyes to green or red light. In addition, the layers
OO A ~ O A O
A 0Q o
000 O A 0
o a AoAoAo
-8- NR,
green light
2 AaQ
crystal with latent
image speck
Sensitivity for:
red light
Fig. 2. General structure of a color negative film ( A silver halide crystals,
0couplers); b blue-sensitive layer, g green-sensitive layer, and r red-sensitive
layer; top: before development, bottom: after development.
dye (magenta)
Fig. 3. Reactions during exposure and development of a color negative film
(example: green-sensitive layer).
In modern color negative films there are some special
couplers besides the standard couplers which fulfill image
processing functions: colored couplers and developmentinhibitor-releasing (DIR) couplers (Fig. 4).
The colored couplers are compounds which change their
color on reaction with the oxidized developert6? In contrast,
in the reaction of a DIR coupler with the oxidized developer,
an inhibitor is released which can freely diffuse in the film in
all direction^^'^. Once adsorbed, the inhibitor contaminates
the latent-image specks (Ag,) and partially developed silver
particles, thus decreasing their catalytic efficiency and retarding the development process.
The different reactions during chemical development of a
color negative film are summarized in Figure 5.
Joachim Lohmann was born in 1939 in Hamburg. After aformal education (1959 to 1966) in
Hamburg and Berlin he obtained his doctorate from the University of Hamburg in 1969 (supervisor H.-F Griitzmacher). He then spent a year carrying out postdoctoral research with Professor
George Porter in London. Since 1971 he has been actively engaged in research at Ada-Gevaert:
at first, in Mortelsen (Belgium), where he was involved in research on silver-free photographic
systems; since 1974 he has been working on photographic filmsand systems in Leverkusen, where
he was appointed research director in 1981.
Angew. Chem. Inl. Ed. Engl. 28 (1989) 1601-1612
colored coupler
DIR coupler
dye (magenta)
dye Imagento)
Fig. 4. Reactions of the oxidized developer with colored couplers and with DIR couplers in a color negative film
cessing results in a modified image function g ( x , y ) , the
recording of which (e.g. by a laser on photographic paper)
yields the improved imageL8].
There are two modes in which the processing can be carcolored coupler
ried out (Fig. 7). In the spatial mode the numbers of the
image function f(x,y ) are changed directly, either individu-
f (x,y)
spatial domain
Fig. 5. Chemical processing of a color negative film
2.2. Digital Image Processing
An analog image must first be digitized to render it accessible to digital processing (Fig. 6). A scanner reads the image
by dividing the image area into discrete picture elements
(pixels), and each element with the spatial coordinates x,y
is assigned a quantized density value f(x,y ) (a pixel in a
color image is assigned three quantized color densities). The
smaller the pixels and the finer the density steps, the more
precisely an image can be digitally described. The image
function f ( x , y ) corresponds to an array of numbers; pro-
f (x,y 1
Fig. 6. Digital image processing.
A n g m Chem Ini. Ed. Engl. 28 (1989) 1601-1612
frequency domain
Fig. 7. Processing modes in digital image processing; x, y spatial coordinates;
u , I' spatial frequency coordinates;f(.r, y), g(x. y ) primary and improved image
function in the spatial domain; F(u, v). G(u, v ) primary and improved image
function in the frequency domain; H,, H,- operators in the spatial and frequency domain.
aIly according to a given scheme (look-up tables) or as a
function of the neighboring numbers. For processing in the
frequency domain, the spatial function f ( x , y ) is first transformed into a two-dimensional frequency function F(u, v )
using a Fourier transform, where u and v are the spatial
frequency coordinates. The low frequencies contain information on luminance and coarse image structure, whereas
the high frequencies contain information on reflectivity,
edges, and granularity. Certain image properties can be enhanced or suppressed by amplification or reduction of the
respective frequency range. A systematic distortion of the
image produced by a moving camera or by an unfocussed
objective lens (perturbation function H) can be eliminated by
applying H*, the inverse of the perturbation
Thus, the frequency mode offers specific routes for image
improvement, and in addition some mathematical operations are very much simplified. After processing the im1603
proved frequency function G (u, v) is transformed back into
the modified spatial function g(x, y ) .
Processing in the frequency domain is of great importance
in digital image processing; frequency-dependent algorithms
also seem to be operative in the visual system (see Section
4.3), but not in the photographic film. Processing in the
frequency domain will therefore be dealt with only very
briefly here.
2.3. Visual System
The human visual system consists of the eye (with its lightsensitive retina) and certain parts of the brain (Fig. 8). In the
right eye
3. Noise (Granularity)
Noise is the random variation of a signal and becomes
more pronounced with increasing amplification.
3.1. Film
In photographic images the phenomenon of noise is
known as granularity-the fluctuation of the optical density
in a uniformly exposed image area on a microscopic scale.
The relation between amplification and noise becomes evident by the fact that high-speed films are generally grainier
and that push-processing (extension of development time to
achieve higher speed) also increases granularity. The source
for the granularity of a film lies in the random position and
the response of the individual silver halide crystals. During
development, the granularity of the silver image is translated
into a granularity of the color image. This transformation,
however, also provides the chance of improving the granularity.
In the presence of DIR couplers inhibitors are released
during the development process which diffuse into the surrounding region and prevent the development of adjacent
crystals; this leads to smaller dye clouds and results in less
granularity (Fig. 10). Decisive in this method is that the sensitivity is not affected, for this is determined by the crystal
size during exposure, which remains constant.
Fig. 8. Human visual system: view of the brain from below [9]
photoreceptor cells of the retina (Fig. 9) the incident light
The phopulses are converted into electrical signals.[”.
toreceptor cells contain about 120 million rods which register only luminance, and about 6 million color-sensitive cones
(three kinds of cone cells with different spectral sensitivities;
absorption maxima at 440, 545 and 580 nrnr6]). The signals
flow via several intermediate cells which are connected crosswise to the ca. one million ganglion cells, and from there to
The complex network of cells in the retina
indicates that image processing functions are carried out not
only in the brain but also in the retina.
Fig. 9. Structure of the retina (taken from [lo])
Fig. 10. Reduction of granularity in a color negative film by inhibitors which
are released from DIR couplers.
This reduction of the granularity by DIR couplers is connected with the chemistry of chromogenic development. To
exploit this effect for black-and-white films, the color
“black” must be produced as a mixture of dyes. This is the
concept of chromogenic black-and-white films.[’41
Furthermore, an incomplete transformation of the silver
image into the dye image is achieved if the oxidized developer
(see Fig. 5 ) is partially consumed by competitive non-chromogenic reactions. Potential competing reactions are the al(the more prokaline hydrolysis of the oxidized
nounced, the slower the coupling reaction), reactions with
incorporated reducing compounds (e.g. dioctylhydroquinone), or reactions with incorporated couplers which produce colorless products.‘’61
In all these cases the yield of dye based on the amount of
silver halide is diminished. The reduced yield of dye must be
Angew. Chem. I n [ . Ed. Engt. 28 (1989) 1601-1612
compensated for by an increase in silver halide coverage. An
improvement of granularity thus has its price.
A completely different approach for decreasing the granularity is to design dyes with a certain diffusibility under the
alkaline conditions of development.'' '1 Such a controlled
diffusion smoothes out local differences in density and leads
to a local density average. Figure 1 1 demonstrates by way of
low deviation of the signal from the local surround the noisy
pixels can be improved selectively by defining thresholds.
Only those pixel values are replaced by the local average
value (i.e. the difference to the local surround is set equal to
zero) whose differences from the local surround are higher or
lower than preset thresholds (Fig. 13).['81 In this way, the
Fig. 11. Reduction of granularity in a color negative film by controlled diffusion of the image dyes.
input signal +
(local difference)
Fig. 13. Digital reduction ofnoise; restriction oflocal averaging to signals with
E , ) or high (> E ~ local
differences from the surround.
low (i
example how the granularity is improved by this method, but
at the same time it reveals the disadvantage of the method,
namely the loss of definition with too much diffusion.
Similar to local averaging in one image, is averaging over
several identical but independently taken images, which is
achieved by exposure through a set of identical negatives.
Owing to its random nature the granularity is reduced, but
the image information is retained.
white points in the lunar image of Figure 14 have been removed. In some cases it is advantageous to use the median of
the surround pixels in place of the algebraic mean.[']
3.2. Digital Processing
In a digitized image, noise is the random fluctuation of the
image function in a uniform field expressed as deviations of
individual pixel values in an otherwise homogeneous field.
Noise can be eliminated in a straightforward manner by
substituting each number by the rounded-off local average
(neighborhood averaging; spatial domain low pass filter). In
the example shown in Figure 12, the local average is calculated as the algebraic mean of a particular number and its eight
nearest neighbors.
Local digital averaging corresponds to averaging by diffusion in a film. In both cases the sharpness is affected. Therefore, rather complicated operators are applied in which adjacent pixels have lower weighting or in which the weighting of
neighboring pixels is variable depending on the local gradiIf the noise is characterized by a particularly high or
6 6 6 6 6
6 6 6 6 6
Averaging over severaI independent images of the same
object decreases granularity in digital images in the same way
as in film images.
Another method having no parallel in film chemistry is
carried out in the frequency domain. Noise is typically encoded in the high spatial frequencies. Filtering out the high
frequencies preferentially suppresses noise (Fig. 15). However, this method also impairs the definition, but it is possible
to minimize this by the use of special f i l t e r ~ . I ~ ~ ]
6 6 6 6 6
6 6 6 6 6
6 6 6 7 6
6 6 6 6 6
Fig. 14. Digital noise reduction by local averaging [ 2 b]
6 6 6 6 6
6 6 6 6 6
6 6 6 6 6
Fig. 12. Digital reduction of noise by local averaging
Angew. Chem. hi.Ed. Engl. 28 (1989) 1601-1612
3.3. Visual System
We rarely experience noise in vision-only at near-darkness may we get an impression of flickering.["* Obviously, the visual system copes very effectively with noise.
Fig. 15. Example ofdigital reduction of noise by application ofa low-pass filter
in the frequency domain (for explanation of symbols see Fig. 7 ) .
Visual noise could have two sources: fluctuation of the
input signal, which will be more pronounced at low light
intensities (photon noise), or spatial-temporal deviations in
the recording and transmission of the primary signals (system noise). The visual system masters intensity differences of
1 : lo1’; the brain receives maximum differences of 1 :102.[201
This shows that the visual system can carefully control amplification, thus minimizing the noise problem.
Suppression of noise occurs in the visual system according
to the same mechanism as applied in chemical and digital
processing: by averaging both in spatial and in temporal
Spatial averaging: there are many more receptor cells than
ganglion cells in the retina; each ganglion cell averages
over approximately 100 receptor cells.
- Temporal averaging: the time resolution of the visual system is rather low; this forms the basis for cinematography.
The eye cannot resolve individual images at frequencies of
20 images s-’ or
but integrates the optical
signals over a certain time span. The extent to which this
improves granularity is immediately clear upon viewing an
individual (still) image of a motion picture film. The temporal averaging is now no longer operative, and the coarse
grain of the individual image becomes apparent.
An additional improvement of the signal-to-noise ratio is
achieved in the visual system by transmitting only those signals to the brain which cross a certain (possibly adjustable)
9 (no e.e.)
9 (e.e.)
distance -+
Fig. 16. Edge reproduction (in analog and digital notation). Top. original signal. Bottom: recorded signal with (e.e.) or without (no e.e.) edge effect.
The more important technique is that of “lateral inhibition’’ with inhibitors released from DIR couplers. Figure 17
4. Sharpness
The sharpness of an image determines the quality of the
reproduction of an edge (Fig. 16). The steeper the profile of
a reproduced edge, the better the sharpness of the system.
Improving sharpness means making an edge profile steeper
or even overshooting the edges. Such an edge-effect improves the impression of sharpness.
distance -+
4.1. Film
The sharpness of a color negative film can be chemically
enhanced by two methods.
Fig. 17. Chemical generation of an edge effect in a film by lateral inhibition
during the development process (schematic). Top: concentration of inhibitor
without diffusion (-)and with diffusion (---). Bottom: corresponding optical
Angen. Chem. Int. Ed. Engl. 28 (1989) t601-16t2
shows the situation at an edge during development in the
presence of a DIR couplers: image dye and inhibitor are
formed on the high density side of the edge (see Section 2.1).
Because of the concentration gradient the inhibitor diffuses
in the direction indicated. That is, in the immediate vicinity
of the edge the inhibitor concentration on the high density
side decreases, while that on the low density side increases.
The degree of development is affected accordingly. In this
way an edge effect is generated. Chemical edge effects have
been described quantitatively.[231
The other method for improving sharpness is “unsharp
masking”, well known to darkroom specialists.[61In this
method a positive, unsharp image is printed in combination
with the negative, sharp original image. Edge effects are
thereby produced in the print. This principle can also be
adapted for use in a film (Fig. 18). In this case a negative,
adjacent numbers. This operation transforms the left string
f(x) into the right string g ( x ) exhibiting an edge effect.
Thus, in digital lateral inhibition, each pixel is influenced in
an inhibitory mode by its neighborhood-in analogy to
chemical lateral inhibition.
Local differences are more or less emphasized depending
on the form of the operator. For example, the operator
[ - 1 2 - I] records differences only; this corresponds to a
(negative) second derivative of the image function and leads
to a contour image.
The calculation of local differences forms the basis of a
whole class of methods for the enhancement of sharpness:
statistical differencing,‘’“. ’1 and use
of second order derivative.[’] The combination of an original
image function with its second order derivative results in an
edge effect, as shown in the top portion of Figure 20.
f ( x J- c.f”(x1
Fig. 18 Chemical generation of an edge effect by unsharp masking in the
green-sensitive double layer of a color negative film (9.g’: green-sensitive partial layers, K : coupler); a) before development, b) after development, c) summation of densities in the partial layers g and g’.
Fig. 20. Digital generation of an edge effect by combination of the original
image functionf(x)- top: with its second order derivativef”(s). bottom: with
a functionJ(x) of reduced sharpness (unsharp masking).
sharp partial image is generated in an upper layer and a
positive unsharp, partial image in a lower layer. The total
image from both partial images shows a clear edge effect.IZ4]
Local lateral inhibition (or local differencing) amplifies
signal differences and is therefore the counterpart of local
averaging. Not only are edges enhanced, but also random
signal differences, i.e. the noise. To overcome this dilemma,
very sophisticated methods are applied or thresholds are
introduced (see also Section 3.2). For example, the operator
can be so modified that only local differences within preset
threshold values and E’ are amplified (Fig. 13), whereas
local differences outside these thresholds are interpreted as
noise and are suppressed.[*,‘*I
In digital “unsharp masking” the original image function
f ( x ) is combined-in complete analogy to the corresponding
chemical method-with a function of reduced sharpness
f(x). Figure 20 (bottom) demonstrates that the resulting
function exhibits an edge
The digital methods for enhancing sharpness by processing in the frequency domain have no parallel in film chemistry. The information on edges is preferentially encoded in the
high spatial frequencies; by amplification of this frequency
range (high frequency emphasize filter) the edges can be emphasized.[2.’. ’]
4.2. Digital Processing
There exist corresponding digital methods to both techniques for chemical enhancement of sharpness.
In digital “lateral inhibition” (center surround filter; high
pass convolution mask[’]) each pixel value is modified, relative to its local surround. Figure 19 gives a simple example
with the operator [ - 1 3 - I] to be applied to each number.
The operator gives the instruction to take the threefold value
of a particular number and to substract the values of both
. 5 5 5 6 6 6. .
f (XI
g (x)
Fig. 19. Digital generation of an edge effect by lateral inhibition.
Angen. Chem. Inr. Ed. Engl. 28 (1989) 1601-1612
4.3. Visual System
An edge effect enhances recognition of details in the visual
system, too. Figure 21 shows stepwise bands of different gray
levels. This image gives the impression that the density in
each band is increasing or decreasing near the edges. In fact,
the density of the bands are completely uniform; the perceived edge effect is purely a visual phenomenon (Mach's
bands 1Z51).
The cause for the edge effects in the visual system is again
lateral inhibition (Fig. 22). The receptor cells send activating
signals (red) to the nearest ganglion cell but send inhibiting
signals (blue) to more distant ganglion cells. Thus, each ganglion cell is subjected to a sum of activating and inhibiting
signals. In complete analogy to chemical or digital lateral
inhibition such a mechanism generates an edge effect.
The effect of lateral inhibition is also evident in the Hermann-effect : on viewing the black squares in Figure 23
ghost-spots are perceived at the intersections of the white
bars between the black squares. At the intersections the lateral inhibition is relatively strong. Here, four inhibiting signals
Fig. 23. Hermann effect; top: visual phenomenon; bottom: explanation.
distance +
Fig. 21. Edge effect as visual phenomenon (Mach's bands).
from the surrounding region act on one activating signal,
whereas in other white areas only two inhibiting signals act
on one activating signal. The stronger lateral inhibition at
the intersections gives the impression of lower brightness.[271
Furthermore, the perception of sharpness in the visual
system is spatial-frequency dependent : The visual resolution
of a line pattern is better at medium spatial frequencies than
at low or high spatial f r e q ~ e n c i e s . [ " ~ ~ ~ ~ ~ * ~
5. Color Reproduction
Fig. 22. Generation of edge effects in the visual system by lateral inhibition in
the retina (from [26]).
Original colors are incorrectly recorded by any imaging
system, e.g., by a film, because the dyes used have non-ideal
absorptions (Fig. 24). When exposed, the color sensors of a
film partially register wrong signals because of their nonideal spectral response; for example, blue light is not only
recorded in the blue-sensitive layer b, but to a certain extent
also in the green-sensitive layer g. During printing the color
information is falsified again, this time by the side-absorptions of the image dyes (the blue light of the printer is modulated not only by the yellow dye in b, but also to a minor
extent by the magenta dye in g).
The situation can be described by a matrix of three equations (Fig. 25).[291The equations contain six interference or
Angen. Chem. In[. Ed. Engl. 28 (1989) 1601-1612
colored coupler
wavelength [nm]
wavelength [nm]
Fig. 24. Incorrect encoding of color signals in a color negative film by nonideal absorption of the dyes used. Top: exposure. Bottom: read-out (printing).
Fig. 26. Improvement of color reproduction in a color negative film by use of
colored couplers; top: absorption of green-sensitive layer g before development; bottom: absorption of g after exposure and development.
correction terms which have different weighting. The mutual
interference of the color signals affects color saturation and
for the increase of the yellow density caused by the side-absorption of the magenta dye. The film exhibits a uniform
yellow density before and after development which is filtered
off in the printing process. Theoretically, all six interference
terms can be corrected for by appropriate colored couplers.
In practice only yellow colored couplers are used for the
green-sensitive layer and red colored couplers for the redsensitive layer. These colored couplers give color negative
films an orange tint.
The other method “vertical inhibition”, is based once
again on the use of DIR couplers and the action of the
inhibitors released during development. If, for example, a
color negative film is exposed to a lot of green light and little
blue light and red light, more magenta and less yellow dye
and red dye are produced (Fig. 27). In the presence of a DIR
coupler more inhibitor is released in layer g than in layers b
and r. A vertical concentration gradient is built up, resulting
in a diffusion of inhibitor from layer g into both adjacent
layers, thus inhibiting development there. The situation can
be arranged in such a way, that the development of the
yellow dye in layer b is inhibited to such an extent that the
undesired yellow side-absorption of the magenta dyeing is
just compensated for. The diffusion of inhibitor out of the
layer g simultaneously enhances the development in this layer. Both effects-the inhibition in the adjacent layers and the
activation in the dominant layer--contribute to increased
color saturation. The inhibitor of one layer acts on the other
two layers (as shown in Fig. 27); incorporation of DIR couplers into all three layers generates a total of six inter-image
effects, which correspond to the six correction terms in the
matrix of Figure 25.
+ k,B
- k,B
- k7B
Fig. 25. Calculation of the original color signals B , G and R from the degraded signals B, G and R; k, are constants for a specific recording system.
5.1. Film
In the case of a film the color signals B,G, and R correspond to the optical densities (determined by the concentration of the respective dyes) as measured behind blue, green,
and red filters.
The chemist has two methods available for improving
(correcting) the color reproduction by interaction of the color signals with each other according to the matrix in Figure
The first method involves the use of colored couplers
(Fig. 26). For example, to correct for the yellow side-absorption of the magenta dye a yellow coupler is additionally
incorporated into the green-sensitive layer; this coupler reacts with loss of its color. At a certain ratio of standard
coupler and colored coupler the decrease in yellow density
achieved by the reaction of the colored coupler compensates
Angeu. Chem. In!. Ed. Engl. 28 (i989) 1601-1612
Another way to digitally influence color reproduction is
by first transforming the input signals R, G and B into a
luminance signal y and two chrominance signals u and v. The
tranformations are given in Figure 28.[21’1 The color saturation is increased if both chrominance signals are equally
amplified. The luminance signal is used to improve noise,
contrast, sharpness and brightness. The modified signals y’,
u‘ and v’ are then converted into improved signals R’, G’ and
B . The increase in color saturation achieved by this method
is shown in Figure 29.
wavelength [nml -i
Fig. 27. Improvement ofcolor reproduction in a color negative film by vertical
inhibition. Top: absorption in the three partial layers of a developed film without vertical inhibition. Bottom: absorption in a film with vertical inhibition.
5.2. Digital Processing
In digital image processing the mutual interference of the
color signals R, G and B can be eliminated by applying the
The six correction terms are specific
matrix of Figure 25.[21a1
for a particular image system (for example for a particular
type of film) which affords the input data for the digital
image processing. After digital matrixing, the color failures
of the original image are corrected with regard to tone and
R’ G’ B’
y= a43 ib.G
re transformation
amplification of
s. v
Fig. 28. Improvement of color saturation by digital amplification of chrominance and of image quality by modification of the luminance.
Fig. 29. Top: conventional print of a slide. Bottom: print of same slide but with
digital processing (digital image processing by CRT printer Agfa-Gevaert AG).
5.3. Visual System
Color results from a non-uniform excitation of the three
kinds of color-sensitive receptor cells.
The visual system has the ability to perceive the colors of
an object as constant regardless of the color of the illuminating light. This has been demonstrated by Landr301in a decisive experiment shown in Figure 30: a color board is illuminated by three light sources (blue, green, red) and the
reflectance striking the eye is measured. In the first case a
certain reflectance R , , G , , B, strikes the eye from the blue
field. In the second case the light sources are adjusted in such
a way that exactly the same reflectance R , , G , , B, now
comes from the formerly yellow-green field. We should expect that this field now appears blue. But this is not so; this
field is seen as yellow-green in both cases, and similarly the
other fields are perceived as unchanged with regard to their
color. Obviously, the eye relates the color signals from any
field to its surroundings. A triplet of relative signals determines the color perceived.
In the context of this comparison of processing algorithms
it is also interesting that in the visual system the signals R, G
Angel$.. Chrm. Inr.
Ed. Engl. 28 11989) 1601-1612
Fig. 30. Experiment for the demonstration of the eye’s constancy of color perceptlon. The values quoted are emisslon
densities [mWsr- ‘m ’1 (from 1301).
Frequency filter
Frequency filters are common in digitdl processing and also
in the visual system.
The correction of color reproduction in color negative
films by matrixing (colored couplers or vertical inhibition)
corresponds to matrixing in digital processing. In both cases,
the R, G, and B color signals are corrected by mutual interaction. The transformation of the R, G and B signals into
luminance and chrominance signals is used in both digitdl
and visual processing.
Some striking analogies occur in the methods of image
enhancement in these vastly different systems--the chemical
film, the electronic computer, the biological visual system.
The methods are written in codes specific to these systems;
these are chemical reactions and diffusion in the film, electronic data processing in the computer, and neural mechanisms in the visual system.
The analogies described for image processing are relevant
for the future of the photographic technology in two ways.
Firstly, studying the similarities of digital and biologicdl
mechanisms for image enhancement may stimulate further
improvement of the photographic film. Secondly, a competing technology becomes evident: A photographic image can
be improved by chemical means incorporated into the film as
well as by digital image processing in the finishing stage.
Future films and their processing may perhaps be simplified
because the computer takes over some processing functions.
Whether chemical, parallel processing is superior to digital
processing or whether the combination of both technologies
will give the best results is still an open question.
Color Constancy
I am very grateful to my colleugue Dip1.-Ing. Eduard
Wugensonnerfor his patience in explaining digital image processing to me.
and B from the cone-shaped cells are combined locally in the
retina in excitatory (+) and inhibitory (-) mode; the composite signals consist of two chrominance signals (R - G;
B - R - G) and one luminance signal (R + G B rod
signal). This model of opponent colors explains many phenomena of color p e r ~ e p t i o n ~321;
~ ’ .the model is similar to the
mechanism of lateral inhibition (in the eye); whereby in both
cases an activating center is surrounded by an inhibiting
+ +
6. Conclusion and Outlook
Table 1 summarizes and compares the methods for image
enhancement discussed thus far. Although some methods are
specific for certain systems, the analogies are evident.
Table 1. Comparison of image enhancing methods
Ag -,Dye
0 Unsharp masking
Colored couplers
- Vertical
Frequency filter
Unsharp masking
Frequency filter
Noise is reduced by local or temporal averaging in all three
image processing systems. Thresholds minimize the adverse
effects of averaging on sharpness; they are applied in both
digital and visual image processing.
Sharpness is improved in all three systems by edge effects
caused by lateral inhibition. The method “unsharp masking”
is used in chemical as well as in digital image processing.
A n g e ~ Chem. I n l . Ed. Engl. 28 (1989) 1601-1612
Received: February 28, 1989 [A 741-IEl
German version: Angew. Chem. 101 (1989) 1633
[l] P. Haberdcker: Digitale Bildverarbeitung, Hanser, Munchen 1987.
121 a ) W K. Pratt: Digital Image Processing, Wiley, New York 1978; b)[2a],
p. 320; c)[2a], p. 323.
131 a) R. C. Gonzdlez, P. Wintz: Digital Image Processmg, Addison-Wesley,
London 1977; b)[3a], p. 199ff; c ) [ 3 a ] , p. 139; d)[3a], p. 153.
[4] E . L. Hall: Compulrr Image Processing and Recognition, Academic Press,
New York 1979.
[5] D. C. C. Wang, A. H. Vagnucci, C. C. Li. Computer Vision. Graphics and
Image Processing 24 (1983) 363.
[6] R. W. G. Hunt: The Reproduction of Colour, Fountain Press, Tolworth
171 G. Haist: Modern Photographic Processing, Vol. 2 , Wiley, New York 1979,
p. 501.
[8] T. M. Cannon, 9. R. Hunt, Sci. Am. 245 (1981) No. 4, p. 136.
191 D. H. Hubel, T. N. Wiesel, Sci. A m . 241 (1979) No. 3, p. 130; Spektrum
Wiss. 1979. No. 11, p. 106.
[lo] T. Poggio, C. Koch, Sci. Am. 256 (1987) No. 5 , p. 42; Spektrum Wiss. 1987,
No. 7, p. 78.
[I 11 T. N. Cornsweet: Visual Perception, Academic Press, New York 1970.
[l2] J. L. Schnepf, D. A. Baylor, Sci. Am. 256 (1987) No. 4, p. 32.
[13] R. H. Masland, Sci. Am. 255 (1986) No. 6, p. 90.
1141 R. Scheerer, E. Ram, J. Lohmann, EP0045427 (1981). Agfa-Gevaert.
[15] a) T. H. James (Ed.): The Theory of the Photographic Process, Macmillan,
New York 1977; b) [15a], p. 343.
[16] W. Piischel, DBP 1099350 (1959), Agfa-Gevaert.
[17] a) [lsa], p. 621; b) R. E. Booms, US-Pat. 4420556 (1981). Eastman Kodak.
1181 R. W. G. Hunt, P h p . Techno/. 16 (1985) 12.
[19] A. Rose: Vision, Plenum Press, New York 1973, p. 45.
[20] E S. Werblin, Sci.Am. 228 (1973) No. 1, p. 71.
[21] a) H. Lang: Farbmerrik und Furbensehen, R. Oldenbourg, Miinchen 1978;
b) [21 a] p. 284; c) 121a] p. 329.
Opt. Soc. Am. A 4
1221 H. B. 9arlow.T. P. Kaushal, M. Hawken, A. J. Parker, .
(1987) 2366.
[23] J. R. Jarvis, J. Phot. Sci. 33 (1985) 212.
[24] M. Fujiwara, EP 0 174873 (1984). Konishiroku.
1251 F. Ratliff, Sci. A m . 226 (1972) No. 6, p. 90.
[261 T. Poggio, Sci. Am. 250 (1984) No. 4, p. 68; Spektrum Wiss.f 9 8 4 , No. 6,
p. 114.
[27] J. P. Frisby: Sehen, Heinz Moos Verlagsges., Giitersloh, p. 156.
I281 F. W. Campbell, L. Maffei, Sci. Am. 231 (1974) No. 5 , p. 106.
[29] P. Kowaliski: Applied Photographic Theory, Wiley, London 1972,
p. 258.
[30] E. H. Land, Sci. A m . 237 (1977) No. 12, p. 108.
[31] M. S. Livingstone, D. H. Hubel, J. Neurosci. 4 (1984) 309.
[32] D. H. Hubel: Auge und Gehirn, Spektrum der Wissenschaft Verlagsges.,
Heidelberg 1989.
Angew. Chem. Int. Ed. Engl. 28 (1989) 1601-1612
Без категории
Размер файла
2 730 Кб
visual, enhancementчchemical, image, digital
Пожаловаться на содержимое документа