Analog and digital • Scanners • Principle •
The Software
RECO Technology is binuscan's exclusive method of color rebuilding and
correction. While other color correction software packages concentrate
on fixed calibrations and the use of Photoshop "transfer" functions,
RECO Technology is a completely different method. Based on Artificial
Intelligence (AI), this means that the software actually intelligently
analyzes each individual image and makes educated decisions about how
to best correct and rebuild it. The following is a brief description of
that process.
Analog and digital
In photolithography, color representation ranges vary from -infinity to
+infinity. In desktop publishing, and in a general manner in every computer-related
image technology, real world data (analog) is reduced to representations
of a limited range, which reduces acuteness in the perception of input
data: this is the digitization process.
Scanners
Digital scanners use a CCD array to acquire images. Considering a scanner
is able to represent one pixel with 8 bits per color channel, the scanning
range will spread from 0 to 255 (2 raised to the 8th power) and the levels
histogram can then be represented as follows:
If a scanner is able to represent one pixel with 16 bits per color channel,
digitizing the same image will give practically the same histogram, but
with a better representation range, as can be seen in the number of levels
of the following histogram:
In principle, the color representation range is wider in 16 bits per pixel
than in 8 bits per pixel, but in reality, the acquisition result depends
more on the quality of the CCD array than on the number of colors used
per pixel. When outputting films, the ranges will be reduced to integer
values ranging from 0 to 100% which will represent the screen dot size.
Acquisition resolutions are measured in dots per inch (dpi). High resolution
images can always be re-sampled to a lower resolution without loss of
quality due to interpolation. Conversely, scanners interpolate (the acquisition
is done at a lower resolution and pixels are generated artificially) when
resolutions other than the scanner's native resolution is desired or when
a scan is performed at a resolution that is higher than the native resolution.
It is important to be aware that the color filters for each image component
(for example, red) always misinterpret a few color particles (for example,
a little bit of green is filtered as red).
The human eye is able to discern densities ranging from 0.10 to 1.80 when
looking at the reflection of an opaque object and from 0.40 to 2.40 when
looking at a transparent object.
When it comes to scanning, while reflective originals always remain in
the 0.10-2.00 range, transparent originals can have a density range anywhere
between 0.00 and 4.00 making them more difficult to acquire and to re-center
in the human eye's perception range. For this reason, scanners must be
able to digitize an image with a density range between 0.00 and 4.00.
The histogram of the input levels reflects this density range. A dark
image with a 2.00-4.00 density range will show a left-sided histogram,
while an over-exposed scan with a 0.00-2.00 density range will have a
right-sided histogram.
Principle
Consider an image digitized on a scanner, with the following histogram:
The acquisition obviously did not result in an even distribution of levels
(with optimal tone mapping as the goal). Plus, data was lost with the
limitation of the representation range due to the digitizing process.
If you try to use a conventional image retouching software to homogeneously
re-map the levels by spreading them, the following histogram will be obtained:
All the software did was lay out the same number of levels obtained at
the acquisition, by creating gaps in between. This way, the color data
and the number of levels used remain identical.
The binuscan technology seeks to use and to re-organize in an optimal
way the histogram's levels. Levels that were not captured at the time
of acquisition are artificially "created" by the RECO technology.
Those excess pixels generated correspond to a smoother transition (therefore
richer in color restitution) inside the image. This way, we can get closer
to the original image, which was limited by the digitizing process. Even
though, to a certain extent, pixel values in the processed image may not
be found in the initial image, this does not modify the original image
since determining pixels are not altered in any way.
With conventional image retouching software, correction operations are
performed sequentially: color saturation, contrast enhancement, unsharp
masking, etc. thus accumulating approximation errors in each successive
calculation. Conversely, the binuscan technology incorporates all the
image correction-related operations in a single operation, minimizing
the error.
The process goes even further in this concept as it takes into account
a great number of information directly related to the image. It does not
only analyze each individual pixel, but also analyzes the surrounding
pixels, the values of the white and black points, the image size and,
of course, its histogram. This way, the correction process is adapted
automatically for each image by modifying its own parameters in order
to obtain an appropriate recalculation of each pixel during the processing
phase. This kind of process is more efficient and accurate than the application
of successive operations with fixed values, whatever the image. For example,
it is obvious that unsharp masking factor and gradation cannot be applied
evenly on different-sized images.
Another strong point of the technology is the improvement of the unsharp
masking and color saturation algorithms used until now.
The binuscan technology also applies common sense rules: since the whole
appreciation is based on the human eye, light areas will be more represented
in the histogram than dark areas. Eventually, the original image used
in our demonstration, once analyzed and processed with binuscan, will
show a resulting histogram as follows:
The Software
Every scanner has at least one weakness (sharpness, prevailing color,
etc.). The binuscan input modules are designed to address and correct
these deviations by forcing parameters during the analysis phase (input
corrections). These corrections are then applied during the processing
phase based upon the input device, still allowing customization for different
output devices. Whichever separation table is used, in every case the
color proof will be the same as the original. binuscan's
"We were surprised at the consistency of the binuscan results,
regardless of which scanner the raw data came from...binuscan's histogram
analysis is obviously doing something smart." — The Seybold
Report on Desktop Publishing.
"Examination of the histogram shows PhotoPerfect is doing something
clever in terms of interpolating tonal information." — MacWeek,
08.26.96
|
