|
|
||||||||
|
|
|
|
|
|
||||
Image Sort
A Simple Method To Sort A Database of Images According To How Closely They Match A Sample
This new technology uses our proprietary VIN technology and
is very powerful. To the best of
our knowledge it offers a new structure for signature recognition not used by
signature recognition companies in the past. Since this change is so different
to anything in use of which we are aware, here is an explanation of the
difference in a few brief sentences.
The most recent technology we invented last year allowed a
signature to be classified by a ten-digit number. This was very powerful since is afforded a way to rapidly
class a signature, and then to compare the VIN (Very Intelligent Number) to
another VIN number through an efficient mathematical operation. The result of
the comparison yielded a number of errors or non-feature matches, which could be
entered into some statistical routines to produce a likelihood of match at
various confidence levels. In our
opinion this was much more attractive than the traditional method of signature
recognition where each feature measured was combined with the history of
measurements of a given feature, statistics derived, and the feature could be
weighted or averaged. The
traditional approach was complicated, and involved storing large amounts of
data. The advantage of the traditional recognition techniques over the newer VIN
methods was that the abnormalities present at any one sample were averaged over
several samples, so could be minimized. In the final analysis both methods had
merit; however, since the VIN method was so simple to deploy and seemed to be as
accurate as any other method, the VIN method was the superior method of the two.
The Image Sort method uses the VIN technology.
You must own a license for the VIN
technology before you can order the Image Sort enhancement. Instead
of assigning a probability, the new technique (called the Image Sort method),
simply stores the ten-digit VIN numbers for a given signer in an array.
Then, when a signer is captured the sample signature is mathematically
compared to the list of VIN numbers in the array, and the array is sorted in the
order of the best match first. We developed our own sort algorithm and it can
sort a database of 100 or so images in a fraction of a second. It has not been
tested against very large image databases.
If the signature is stored in a flat file or database
element and named as “(VIN number)”. bmp then the signature VIN number that
most closely matches the sample can be presented side by side with the sample.
The percentage of likelihood of match can be displayed along with an
image of the closest match to the sample image.
In our testing, we found if we stored four or five samples of the
signature, and then sorted these images in the order that they matched the
sample, we achieved remarkable results. We
also tested the process with a large database of images, some which matched and
some which did not. In our test
about 1 of 8 images in the database of all images matched the sample.
We discovered using a small database of 78 images, our method was able to
ALWAYS select the correct signature, and more importantly always had a high
degree of confidence. If you
combine this with a way to visually present the stored image which best matches
the sample, then we believe this is a very powerful method indeed.
Click here to Review Test Results
The traditional VIN number technique offered by our
original ActiveX component simply XOR’d a sample VIN with the master VIN which
then produced a percent confidence the images were the same.
Now, we can see that even though the percent model will work, the signature/image match algorithms illustrated here are much more powerful. For example the document enclosed shows the system picked my signature out of a group of 78 random images EVERY TIME, and it picked ALL instances of my signature before it started picking less likely images. The least likely image (choice # 78) with a confidence of 0 is shown for reference.
It should be noted the original ActiveX component was
designed to allow image transfers without any disk access, there may be a
limitation on the new technology in this regard.
Somehow the VIN numbers and ideally the images of the signatures should
be stored. The VIN numbers are
small (ten digits) so could easily be placed in a SQL Server, Oracle or Access
server. Even My SQL would work.
The signature images could be placed in a flat file on a server.
We use memory resident arrays to sort, and only use disk accesses to store the resultant bitmap signature images. This means in a browser application, the images could be stored in a database such as SQL Server or Oracle. We use this technique to store signature images at large banks. Since the images are typically about 500 - 700 bytes compressed, it works well.
Contact Us