Image analysis – Image enhancement or restoration – Artifact removal or suppression
Reexamination Certificate
1999-03-09
2002-05-21
Rogers, Scott (Department: 2624)
Image analysis
Image enhancement or restoration
Artifact removal or suppression
C382S260000, C382S263000
Reexamination Certificate
active
06393160
ABSTRACT:
TECHNICAL FIELD OF INVENTION
This invention relates to electronic scanning of images, and more particularly to the scanning of photographic prints by reflected light and the removal of surface defects.
BACKGROUND OF THE INVENTION
The present invention is an improvement on a method of correcting defects in a film image using infrared light as taught in U.S. Pat. No. 5,266,805 issued to Albert Edgar, the present inventor. The underlying physics enabling this method is illustrated in FIG.
1
. In
FIG. 1
it is noted that with any color of visible light, such as green light, one or more dyes in a color film absorb light with corresponding low transmission of the light; however, in the infrared wavelength range, the common image forming dyes have a very high transmission approaching 100%, and therefore have little or no effect on transmitted infrared light. On the other hand, most surface defects, such as scratches, fingerprints, or dust particles, degrade the image by refracting light from the optical path. This refraction induced transmission loss is nearly the same in the infrared as it is in the visible, as illustrated in FIG.
1
.
Continuing now with
FIG. 2
, a film substrate
201
has embedded in it a dye layer
202
. Infrared light
204
(
FIG. 2
a
) impinging on the film
201
will pass through the film and emerge as light
206
with nearly 100% transmission because the dye
202
does not absorb infrared light. Conversely, visible light
208
(
FIG. 2
b
) will be absorbed by the dye
202
. If the dye density is selected for a 25% transmission, then 25% of the visible light
210
will be transmitted by the film
201
.
Now assume the film is scratched with a notch
214
(
FIG. 2
c
) such that 20% of the light will be refracted from the optical path before penetrating into the film
201
. When a beam of infrared light
216
strikes the film
201
, 20% will be diverted due to the notch
214
, and a beam of 80% of the infrared light
218
will be transmitted. Finally, let a beam of visible light
220
(
FIG. 2
d
) impinge on the film
201
. Again 20% of the light
222
is diverted by the notch
214
, leaving 80% of the visible light to penetrate the film
201
. However, the dye layer
202
absorbs 75% of that 80%, leaving only 25% of 80%, or 20% of the original light
224
, to pass through the film
201
.
In general, the beam left undiverted by the defect is further divided by dye absorption. In visible light, that absorption represents the desired image, but in infrared that dye absorption is virtually zero. Thus, by dividing the visible light actually transmitted for each pixel by the infrared light actually transmitted, the effect of the defect is divided out, just like division by a norming control experiment, and the defect is thereby corrected. This division process is further clarified in FIG.
3
. The value of a pixel
302
of a visible light image
304
is divided with operator
306
by the value of the corresponding pixel
308
of the infrared light image
310
. The resultant value is placed into pixel
312
of the corrected image
314
. Typically, the process is repeated with visible image
304
received under blue light, then green light, then red light to generate three corrected images representing the blue, green, and red channels of the image
304
.
FIG. 4
is similar to
FIG. 3
in that it shows a process for removing the effect of defects from a visible light image
404
using an infrared light image
406
. Although the operator
408
in
FIG. 4
is a subtraction,
FIG. 4
is mathematically identical to
FIG. 3
because the same result is obtained either by dividing two numbers, or by taking the logarithm of each, subtracting the two values in the logarithmic space, then taking the inverse logarithm of the result. However, the arrangement of
FIG. 4
enables many additional useful functions because within the dotted line
402
, the signals from images
404
and
406
may be split and recombined with a variety of linear functions that would not be possible with the nonlinear processing using the division operator of FIG.
3
.
For example, in
FIG. 5
a visible image
502
and an infrared image
504
are processed by logarithmic function blocks
506
and
508
, respectively, to enter the linear processing dotted block
510
equivalent to block
402
of FIG.
4
. After processing within block
510
is completed, the antilog is taken at function block
512
to produce the corrected image
514
.
Internal to linear processing block
510
, the logarithmic versions of the visible and infrared images are divided into high pass and low pass images with function blocks
520
,
522
,
524
, and
526
. These function blocks are selected such that when the output of the high and low pass blocks are added, the original input results. Further, the high pass function blocks
522
and
526
are equal, and the low pass function blocks
520
and
524
are equal. Under these assumptions, and under the further temporary assumption that the gain block
530
is unity, the topology in linear block
510
produces a result identical to the single subtraction element
408
for FIG.
4
.
Without the logarithmic function blocks
506
,
508
, and
512
, the split frequency topology shown in block
510
would not work. The output of a high pass filter, such as blocks
522
and
526
, averages zero because any sustained bias away from zero is a low frequency that is filtered out in a high frequency block. A signal that averages to zero in small regions obviously passes through zero within those small regions. If function block
540
were a division, as would be required without the logarithmic operators, then the high pass visible signal
542
would often be divided by the zero values as the high pass infrared signal
544
passed through zero, resulting in an infinite high pass corrected signal
546
, which obviously would give erroneous results. However, as configured with block
540
as a subtraction, the process is seen to avoid this problem.
The split frequency topology of
FIG. 5
appears to be a complicated way to produce a mathematically equal result to that produced by the simple topology of FIG.
3
and FIG.
4
. However, by separating the high frequencies as shown in
FIG. 5
, it is possible to overcome limitations in the scanner system by now allowing the gain block
530
to vary from unity. A typical scanner will resolve less detail in infrared light than in visible light. By letting gain block
530
have a value greater than unity, this deficiency can be controlled and corrected.
Often, however, the smudging of detail by a scanner in the infrared region relative to the visible region will vary across the image with focus shifts or the nature of each defect. By allowing the gain block
530
to vary with each section of the image, a much better correction is obtained. In particular, the value of gain is selected such that after subtraction with function block
540
, the resulting high frequency signal
546
is as uncorrelated to the high frequency defect signal
548
as possible. If given the task, a human operator would subtract more or less of the defect signal
548
as controlled by turning the “knob” of gain block
530
. The human operator would stop when the defect “disappears” from corrected image signal
546
as seen by viewing the corrected image
514
. This point is noted by the human operator as “disappearance” of the defect and is mathematically defined as the point at which the defect signal
544
or
548
and the corrected signal
546
are uncorrelated. This process could be repeated for each segment of the image with slightly different values of gain resulting as the optimum gain for each segment.
Despite the flexibility introduced by the gain block
530
of
FIG. 5
, it has been found that often a defect is incompletely nulled because deficiencies in the scanner cause the defect to look different in the infrared and the visible, such that no setting of gain can eliminate all aspects of the defect.
A need has thus arisen for an improved method for image defect corre
Applied Science Fiction
Dinsmore & Shohl LLP
Rogers Scott
LandOfFree
Image defect correction in transform space does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Image defect correction in transform space, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Image defect correction in transform space will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2863668