human-like segementation, elaborate filing
systems are used to organize paper, but it is very difficult (if not
impossible) to keep track of thousands of pieces of paper as they
make their way around the corporate circuit.
When paper is lost, the company (if not
the officers) incur liability. For example, the IRS will probably
not allow the deduction.
segmenter can play a central role in mitigating these issues.
On the image quality side, the storage
method must be able to retain all necessary information on the
document. To paraphrase an eminent philosopher, a system with too
many exceptions is of little use. We need to handle all the pieces
of paper. For example, a dental office might need to put paper
notes, dental x-rays, check copies, and several other categories of
documents into a single system. If we can't get the information into
the computer, we have to continue to rely on the paper files.
We are not advocating throwing the paper
away. Everyone will want to take a few years to get comfortable with
the electronic image first.
images have a number of advantages over paper images. People will
choose the electronic file over the paper file for various reasons.
For example, to use a paper file, people usually have to get out of
their chair (it could be considered work). On the electronic side,
we do not have to move body mass to get what we want.
Electronic documents are easier to copy
and distribute. By using bandwidth instead of the Post Office, we
can afford to ham it up. We don't have to purchase envelopes,
stamps, or paper.
Our colleagues could
look at one of our files without bothering us, and we can set it up
so they don't lose it.
To make all of
this possible and practical, the segmentation must be at human level
(or better), and the compression must take the 25 MByte file down to about 10 KBytes.
Even if this system was just used to store and distribute critical
tax and corporate documents that everyone seems to need to see on a
regular basis, it would be worth its weight in gold. Of course, it
can be applied to a much broader scope.
When we take business trips, we could simply take a photos of a
receipts. Then when our luggage suffers airline abuse that causes
the shampoo bottle to explode and destroy all those receipts (which
probably roughly equal a mortgage payment on our house), we could
still electronically file our expense report with all the necessary
receipts long before we ever get back to the office.
Once the receipts are electonically
attached to the reimbursement record, the IRS auditor can simply
click on the reciept to verify its existence. With paper records,
the filing clerk has to take the list of requests from the auditor,
pull the files, run the copies, refile the records, and send them to
the auditor. There can be some number of iterations.
IRS and other audits are typical of the
use of files, but obviously, the files are used for many other
reasons as well.
Computer records have
proven to be much more durable, distributable, and trackable than
paper documents. With computer records, it is easier and cheaper to
establish, manage, and maintain the nearly all important "paper"
As an example, when the World
Trade Center catastrophe occurred, nearly all the paper records were
lost, but most of the computer records survived.
pictures are highly complex and high quality is desired, most
companies use a transform compressor (usually JPEG) that throws out various colors
as determined by a mathematical algorithm. The output is considered
high quality, but both contrast and detail are compromised
among other things.
TIFF G4 is the workhorse of the document imaging industry. Threshold segmenation is used to
produce bi-tonal images.
The documents should be clean, and any
handwriting could be lost in the process. Fine text may not be
This system usually
reduces one page documents from 25 MBytes to 30 to 50 KBytes.
With TIFF G4 technology, the file sizes are too large and the image
quality is too low to meet the needs of the typical office
|| To be
practical, the same document imaging system needs to be
used on all the documents an office has.
Therefore the document imaging system required in a typical office
needs to reproduce all the information on the document that humans
can see. If segmentation is used at this high of a
quality, the number of blobs will be very high which makes
useable compression harder to obtain.
theory, a geometrical compressor could only
obtain the required compression if the blobs were repeatable. The
blobs are not repeatable when segmentation is taken to human-like
quality. The trade off between segmentation and compression seems to
result in a catch 22 for today's equipment, but to paraphrase
classical military strategy, the worst part of the problem contains
the key to the solution.
When we create
documents on a computer, there is usually extreme repeatability and
very few colors. For example, the word "repeatability" has two
colors, two 'e's, two 'a's, two 't's and two 'i's. There are
similarities and repetition between indiviual letters. For example,
the bottom of the 'i' is similar to the bottom of the 'l'. Nature,
on the other hand, is hardly ever repeatible either in form or
color. For example, every snowflake and fingerprint are different.
In fact, nature almost never does exactly the same thing twice.
A geometrical compressor would have a
field day on human created documents before they were printed out,
scanned in, crumpled up, written on, and otherwise mutilated and
distorted (entropy attacks repeatability and symmetry). The typical
office paper was created by humans and distorted by entropic
processes. If we could reverse the entropic distortions of our
documents, we could recover the original (and repeatable) data they
Since the documents were made by
a human, we know, that except for some photographs, they were
originally very repeatable.
distortion follows rules (or laws). For example, optical aberrations
can be precisely simulated in a known environment. In the typical
office, the environment is not known, but it can be approximated.
Of course, there are a number of
different ways a document can be compromised, but we can estimate
how the more common distortions would occur.
If we use image restoration procedures to
correct the more common distortions, we can pull the image back to
something closer to the original. We can not get the original back
perfect with image restoration, because we have to approximate the
environment of distortion. Then by allowing a tolerance around what
is expected, we can snap the image back to nearly exactly the
With these techniques
we do not recover the original image exactly, but we would recover
the information we would draw from the image.
Therefore, human-like segmentation
combined with geometrical compression can meet the needs of the
typical office, if image restoration is carried to within the
tolerance of an expected artifact.
solution would be implemented in two steps. In the first step, we
would segment and restore. In the second step, we would apply a
geometrical compressor that tolerates acceptable pixel misplacements
as illustrated here.
implementation is complicated and computer intensive, but the
compression could easily be 100 times greater than any conventional
method. Furthermore, humans would not fault the quality.
The technique of combining advanced
segmentation, image restoration, geometrical compression, and
tolerating acceptable artifact distortions will likely have a
dramatic impact on the document handling industry. High quality and
extremely high compression should propel the industry into many new
markets while simultaneously expanding its existing markets.