Image analysis – Applications – Personnel identification
Reexamination Certificate
1999-12-02
2002-07-23
Boudreau, Leo (Department: 2621)
Image analysis
Applications
Personnel identification
Reexamination Certificate
active
06424728
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a computerized system for automatically authenticating signatures. More particularly, it relates to a system utilizing forensic methods of analysis while fully automating signature verification thereby reducing the time necessary for verification of a signature to a fraction of a second using ordinary personal computers.
2. Background Information
American businesses and households write over 60 billion checks per year. Banks and their customers lose nearly ten billion dollars (Bank Administration Institute reports) to check fraud due to shortcomings during processing of those checks.
The Office of the Comptroller of the Currency (OCC) has indicated that check fraud is one of the largest challenges facing financial institutions. Modern technology allows more accurate forgeries to occur, making detection more difficult. In addition, technology allows more realistic counterfeit checks and false identification that can be used in combination with forged signatures to defraud banks.
Many difficulties have been encountered in applying automatic signature verification systems to current data environments within banks. There is a need to verify signatures on checks that have a low resolution or poor quality image. The time it takes to verify each check with current automatic computer programs is far too long.
The OCC advises banks to review checks ensuring that the handwriting or print styles are consistent and that there are no signs of erasure or alteration. Banks should also compare the endorsement signatures on items presented and compare the appearance of the presenter with the signature and picture on their identification.
One commercial automatic signature verification system used by banks is provided by SOFTPRO. This system is mainly based upon a neural network training approach. The verification decision is performed using electronic comparison of images in which the set of parameters of the document's signature is compared with all parameter sets of the master signature of the account. Furthermore, every signature is assigned to one of the six defined agreement categories, AA to F, which can be subdivided into five categories if needed. The basis of both training and decision making (accepting the document signature as genuine or rejecting it as a forgery) are drastically different from those of the present invention.
A bank or other financial institution may process millions of checks each day. Even with the help of computerized visual verification systems, only a small portion of the checks will have their signatures examined. This situation makes automatic verification a necessity to deal with large numbers of checks. The processing speed of a system becomes a critical factor for two reasons; (1) the time period allowed for examining all of the checks is only two to four hours; and (2) the number of computers required to perform the work increases in proportion to the time t takes to verify each check.
The SOFTPRO system discussed previously has a speed of 1,800 documents per hour, using a Pentium II processor, which translates into 90,000 documents per hour using fifty computers. This number is far below the requirement of a large bank.
Another critical problem that has heretofore not been solved deals with the current database used by a bank. Some banks currently scan checks using hand scanners which have low resolution capabilities and poor quality signature card images. For example, documents scanned at 120 dpi and 80 dpi are available by the millions. Rescanning all of those documents is too time consuming and expensive. Current systems recommend a minimum of 200 dpi for scanned document data, and may be capable of using data down to 150 dpi with a greatly reduced accuracy. The current systems with resolutions fail to provide meaningful results if data below 150 dpi are used.
A neural network trained using English style signatures can not be used to analyze Chinese style signatures. The network must be trained to handle type of writing, making it difficult to apply generalized software to different languages. Accordingly, there is a need for a system that can overcome the variations in signature.
SUMMARY OF THE INVENTION
In contrast to computerized systems that give the decision via comparison with only one genuine signature sample, the mechanism of giving the decision in the present invention is consistent with forensic examination of handwriting samples. It is a well established fact in forensic examination that a person cannot write their signature the same way twice. Every sample is different from the other to some degree in that the values of the features of each signature vary as to relative size and two dimensional position, relative slants, curvature of letters, and the like. If the differences in the values of selected features of a target signature lie in a selected range of acceptance, as determined by the natural variations of the signature of the specific person, the signature is accepted as genuine, otherwise, it is classified as an attempted forgery. In this context, the present invention uses image processing and pattern recognition techniques to implement forensic examination concepts during the decision making process by comparing the selected features of a target signature with a reference knowledge, or values, of the signature of the specific person obtained from a set of training or genuine signature samples. Training is done on two levels: global, which includes the entirety of a genuine signature database of many persons, and personal, which includes a set of genuine signature samples of a specific person. While the present invention uses a set of genuine training signature samples for every person, it starts making decisions using only one reference signature sample using what is called here accelerated learning. In the beginning of the learning process, it uses the global knowledge as a starting point to give a decision using only one reference signature. When new genuine signatures become available (from new checks or the like) the reference knowledge of a person's signature is updated until it becomes fully dependent on the genuine signatures of that person after only six genuine samples.
A system for automatically verifying signatures includes a program running on a personal computer using at least one authentic signature that has been scanned into a genuine signature data base. A program caused the personal computer to run various algorithms to clean a digitized image of a target signature. The program then normalizes the image and makes Euclidian weighted measurements of forensic features of the target signature, such features being compared with those of the authentic signature features.
The algorithms used can use authentic signatures having a resolution at least as low as eighty dpi.
REFERENCES:
patent: 4028674 (1977-06-01), Chuang
patent: 5042073 (1991-08-01), Collet et al.
patent: 5251265 (1993-10-01), Dohle et al.
patent: 5559895 (1996-09-01), Lee et al.
Computer Recognition and Human Production of Handwriting; Maan Ammar, Yuuji Yoshida and Teruo Fukumur. World Scientific Publ. Co., 1989, pp. 61-76.
Mainichi; Nagoya University Foreign Student Association Jun. 1989.
Presentation titled:Portable Software for Signature Verification and Analysis(SIGVA1.0),Usable With IBM-PC Compatible Machinesby Maan Ammar, Mohammed Aita, Basil Younaki and Bassam Takwa at Basic and Applied Issues in Handwriting and Drawing Research Conference. University of Western Ontario London, Ontario, Canada Aug. 6-10, 1995.
International Journal of pattern Recognition and Artificial Intelligence “Progress in Verification of Skillfully Simulated Handwritten Signatures” Maan Ammar. World Scientific Publ. Co., 1991, vol. 5, No. 1 & 2, pp. 337-351.
Boudreau Leo
Handal & Morofsky
Werner Brian P.
LandOfFree
Method and apparatus for verification of signatures does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for verification of signatures, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for verification of signatures will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2893971