|
|
Manual Edge Detection Segmentation
|
Like actors in a play, images
have a variety of moods. The
Pac-n-Zoom® user will want to
change the settings to produce
the correct effect. For the most
part, the settings allow the
number of colors to be used. For
example, the setting for an ever
generous rose would allow for
many more colors than the
setting for hard and cold prose.
There are three ways to set the
settings. The first method is
through the
GUI (Graphical
User Interface - the normal program
interface) and the pull down
menus. The second method uses
macros through
the
CLI (Command
Line Interface - such as a DOS
prompt). The CLI allows the
settings to be changed when
running the in the
batch mode.
The third method is to edit the
"Pac-n-Zoom.cfg" with a
text editor.
When Pac-n-Zoom is not running, the settings are
stored in "Pac-n-Zoom.cfg" under the
segment name of
"VIDEO FILTER PARAMETERS". A typical segment
of data might look like the following.
|
|
# VIDEO FILTER PARAMETERS
75 ;
Tier one setting
75 ;
Tier two setting
0 ;
Aggegate pixel size
1 ;
Color sweep setting
150 ;
Orphan pixel merging color
1 ;
Eliminate transitional blobs
75 ;
Near color elimination
7 ;
Number of colors allowed
*
|
These settings are typical
settings for a 7 color (in other
words 7 original and final
colors - the number of scanned
colors could be many thousand)
document.
The following settings are
listed in their order of
importance. The settings at the
top of the list are the most
important.
|
|
Tier One
|
The first step in
extracting information
from visual data is to find the
borders that
have the highest
contrast
color
transitions in
the picture. In Pac-n-Zoom, these
are known as tier one borders.
Tier one borders usually demarcate
the primary artifacts of the
picture.
The tier one setting should be set
low enough to obtain all the
critical information in an image.
For example, when the tier one is
set too high, parts of a letter
may be missing.
When the tier one setting is set
too low, the same letter can be
set to several different shades or
gradual changes can become a
series of steps.
Use the following steps to set the
tier one level.
|
|
|
1.
|
Select Option: Set the
display tier one border option.
|
|
2.
|
Reduce Level: Reduce the tier
one setting to where the
image is segmented into too
many blobs, and note the
low setting (LS).
|
|
3.
|
Increase Level: Increase the
tier one setting until
important demarcations are
lost, and note the high
setting (HS).
|
|
4.
|
Calculate Level: Set the
tier one level with the
following calculation.
Tier 1 Level = (HS + LS) / 2
|
|
Click here to
see some examples of correct and incorrect settings.
The input device usually causes
image degradation along high
contrast borders. Pac-n-Zoom
restores the tier one borders to
closer approximation of the
original image.
|
|
Aggregate Pixel
|
Nearly every computer image,
except the few painted ones, was
captured by
wave front optics focusing
an image onto
charge coupled semiconductors.
Any number of distortions (some
of them are mentioned above)
append themselves to the image.
In particular, a high contrast
color
transition will
tend to scatter hundreds of
orphans.
The optics blur the image. The
depth of focus problems
are particularly destructive,
but
color level
flutter and
optical chromatic aberrations add
their fair share of damage.
Besides optical and transition
distortions there are a range of
frequency and spatial problems
that include
Nyquist noises
and
Johnson aliases.
These are three areas where the
image is compromised but there are
at least several more.
A single
pixel on
these captured
images won't mean much, unless it
constellates with its nearest
neighbors. The aggregate pixel
setting is an abstract setting
that partially determines the
number of pixels needed to make an
artifact significant.
Pac-n-Zoom uses aggregate pixel
setting in several places, and
several problems can occur if it
is not set correctly. There are,
however, only two aggregate pixel
settings (0 and 2).
|
|
|
0:
|
This setting should be used
on all pictures except
video. Its size is 2 by 2
pixels.
|
|
2:
|
Should be used on video
pictures. Its size is 4
pixels wide by 2 pixels high.
|
|
|
Tier Two
|
A gradual change in color will
not trigger a tier one artifact,
but over a number of pixels, the
color can change significantly.
To insure the fidelity of the
image, the color should not be
allowed to change within the
range of human recognition
without starting a new
blob. The
tier two borders serve this
purpose.
Most humans can see the
difference between 4,096 and
32,768 colors, but only about
18% of the people we tested
could see the difference between
32,768 and 262,144 (6 bits for
each
primary color
when using 3 primary colors) colors.
The tier two setting does not
set the amount of the actual
jump (change in color) between
adjacent blobs. It sets the
amount the original picture is
allowed to change inside what will
become one blob (remember: a blob
is all exactly the same color;
the change inside the blob
refers to the amount of change
allowed in the underlying pixel
that will be grouped into the
exact color of the blob). For
example, a tier 2 setting of 9
means that the extreme
individual pixel colors inside a
blob are within 9 color levels
of each other. Therefore, tier 2
blobs insure the fidelity of the
image.
When two adjacent blobs have a
visible color jump, a
contour line forms
(
click here to
see some examples of contour
lines).
When tier two settings are set too
high, the contour lines will stand
out more than desired.
The following steps can be used to
set the tier two settings.
|
|
|
1.
|
Tier One: The tier
two blobs are layed inside
the tier one blobs.
Therefore, the tier one
setting must be set first.
|
|
2.
|
Select Options: Set the
display tier one and display
tier two options.
|
|
3.
|
Set High: Increase the
tier two setting until the
contour lines are too
visible.
|
|
4.
|
Decrease Level: Decrease
the tier two setting by
about 30% (this is an
arbitrary amount). For
example, if the high setting
(from step 3) was 16, the
nominal setting would be 11.
|
|
|
Compatible Colors
|
The tier two sets the amount of
fidelity to the original color. It
does not specificallly set the
amount of jump between two
terraces.
For example, suppose the color
level on adjacent pixels were
ramping as shown in the chart. If
a tier two
blob started
at
pixel number
one and if the tier two blob
setting was 9, a new blob would
start at pixel 4 (because the 69
on pixel 2 has to be in a
different tier 2 than the 79 on
pixel 4).
|
|
| Pixel Number |
1 |
2 |
3 |
4 |
5 |
6 |
|---|
| Color Level |
76 |
69 |
77 |
79 |
72 |
74 |
| Average Color
| 74 |
75 |
|
Therefore the color is terraced to
jump one pixel. If the compatible
color setting was set to 1 or
larger, the two blobs in the table
above would be merged together.
They will also be merged with
other colors throughout the image
that are within the tolerance set
by the setting.
Fewer colors will compress
further, but when taken too far,
the image will begin to look
coarse (like a cartoon instead of
a photograph).
From our experience, the
following are appropriate
compatible color settings in
color levels.
|
|
|
1.
|
Photographs - 9
|
|
2.
|
Video - 15
|
|
3.
|
Line Art - 150
|
|
|
Eliminate Transitional Blobs
|
In the perfect world, if the
aggregate pixel setting is set
accurately, the color should not
be changing for more than the
aggregate pixel length. This
theoretical ideal is often
violated from various
fluctuating distortions (in
other words, the distortion in
different parts of the picture
are different) such as
depth of field issues.
For example, the depth of field
can be dramatically different
between the background and
foreground of an image. Therefore,
in the real world, the color
transitions (the
amount of space it take the
color to change from one level
to the next) can exceed the
aggregate pixel length.
Whenever transitional colors
extend beyond the aggregate pixel
length, entire
blobs can
be transitional.
Even when the documents are
scanned in, it may be desirable
to eliminate transitional blobs.
For example, if there are almost
any
color bleedings from
the printing process, they will
result in transitional blobs.
A transitional blob is defined
as one where the maximum height
or width is either the aggregate
pixel size or smaller, and all
colors transition in reference
to the tier two setting. When we
say "references the tier two
setting," we mean the color
changes more than the tier two
settings. The three possibilities
are plateaus, ramps, and
extremities (
click here to
see them).
Plateaus and ramps are
considered transitional blobs,
but if any color is an
extremity, the blob is not
considered transitional. This
logic removes the transitions
that lie between major blobs
without deleting small amounts
of data that doesn't transition.
The removal of transitional
blobs takes some of the softness
out of a picture. The
transitions will tend to be more
abrupt (higher
contrast). Greater
contrast is usually appreciated
in text documents and line art
but depreciated in photographs
and video.
When transitional blobs are
removed, spacial (
geometric and
statistical) compressors
will have higher compression
ratios.
The eliminate transitional blob
setting has two valid values.
|
|
0:
|
The transitional
blobs will not be
eliminated.
|
|
1:
|
The transitional
blobs will be eliminated.
|
|
|
Eliminate Extra Colors
|
The Pac-n-Zoom segmenter allows
the user to specify the maximum
number of colors in the picture.
The colors with the highest
pixel count
are left in the picture. A first
with the most system is used to
break ties (the first colors
found, starting from the upper
left hand corner then going right
before down, are taken over
later colors).
The elimination of extra colors
deletes color and geometric
detail from a picture. Since the
image restoration is
not perfect, most of this detail
is from sensing distortions.
Users with photographs will
probably want to set this
setting at about 80,000 for the
Internet, and at 250,000 for
professional photography. As of
this writing, we are not entirely
certain of the effects, and
settings greater than 250,000
might be desired in some cases.
Video users will probably want
to set the setting at about
70,000 (this is a guess). For
streaming video, values of
12,000 might be in order
(another guess).
In document handling and line
art, the number of colors is
often minimal (usually between 2
and 5). Therefore the setting is
obviously set to the number of
shades in the picture.
When extra colors are removed,
spacial (geometric and
statistical) compressors will
have higher compression ratios.
|
|
Eliminating Orphan Pixels
|
This setting determines the
amount of
contrast necessary
to leave an
orphan
pixel in
the image. On the
extraction side,
it could be argued that orphans
are worthless and should always
be eliminated regardless of the
amount of contrast. To implement
this opinion, the setting should
be set at 255.
For the most part, orphans don't
help photographs or video and
hurt line art and text documents.
Orphans will lower the compression
ratios of all compressors with the
possible exception of
transform compressors Furthermore,
orphans will prevent
raster
to
vector conversion
in Pac-n-Zoom's
data-tagger. Therefore,
orphans should usually be
eliminated.
The orphan pixel setting has the
following meaning.
|
|
0:
|
All ophans will be
accepted.
|
|
140:
|
Orphans that have a color
contrast of more than 140
on any
primary color will
be accepted. The number,
"140," can be any number
between or including 0 to
255.
|
|
255:
|
No orphans will be allowed.
|
|
|
Sweeping Colors
|
This setting selects the
image restoration algorithm.
We have tried a couple of
setting, and we like the current
one best.
Different distortions types
should theoretically use
different algorithms.
The current setting needs to be
set to one.
|
|
|
|
|
|
© 2004 - 2011 Accelerated I/O, Inc, All
rights reserved.
|
|