Image analysis – Applications – Mail processing
Reexamination Certificate
1995-04-12
2001-07-31
Couso, Jose L. (Department: 2624)
Image analysis
Applications
Mail processing
C382S102000
Reexamination Certificate
active
06269171
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention relates to automatic mail processing and more particularly to a method of exploiting mail stream statistics to improve optical character recognition.
In the United States a large and always growing volume of mail is processed on a daily basis. Although recent hardware and software advances have been made in optical character recognition (OCR) and these advances have improved overall mail throughput, further improvements are desirable in attempting to achieve the economic benefits that would flow from a complete and fully automated bar-coding system.
In conventional OCR methods for processing letter mail and assigning a bar code, an address block location must first be found. Next, the address is processed by a segmentation function whose ultimate goal is to separate each line into individual characters. The recognition process then attempts to identify each pertinent character. If a zip code is read incorrectly and cannot be verified with a database search, a bar code cannot be assigned and manual processing is typically required.
Problems that occur in current address interpretation methods are that they either assign an incorrect zip code or they do not assign a zip code at all. The first problem occurs when a word break is not present at the start or end of the zip code, or a word break has been placed in the middle of a zip code. The second problem occurs when one or more of the correct digits of the zip code are not ranked as the first choice by the recognition process and are therefore not selected.
While statistical analyses focusing on individual mail pieces have been done, the statistics of typical mail streams has not been exploited.
It is an object of the present invention to provide an automated mail processing method which reduces the amount of mail which must be manually processed.
It is another object of the present invention to provide an automated mail processing method which takes advantage of the statistics of the mail stream being processed to improve OCR recognition rates.
SUMMARY OF THE INVENTION
In one aspect of the present invention a method of performing adaptive optical character recognition using correlated mail stream data is provided. Data from mail processing equipment is collected for generating a statistical information database. A decision threshold is determined based on the statistics of the mail stream for assigning characters in a previously rejected mail piece based on correlation statistics. Previously unassigned characters are identified according to the design threshold determination and assignment criteria.
In another aspect of the present invention an optical character recognition method for determining the zip codes on mail pieces is provided. The last line of the address block is searched for the most popular three digit zip codes. A list is made of all the popular three digit codes that are found. The characters that make up the three digit zip codes are chosen regardless of character ranking, word breaks, and character confidence. The candidates are then ranked based on the sum of the character confidences for the individual characters of the three digit zip codes. The three digit zip code with the highest confidence is then assumed to be the correct choice for that image.
In still another aspect of the present invention an optical character recognition method for more efficiently locating an address block in an image using adaptive techniques is provided. An example where the adaptive technique is useful is in locating the address block in mail that originates from large mailings (bills, advertisements, etc.), since the address block location will be the same for images that originate from the same large mailer. Similar images are grouped together based on a compressed form of the image. This is done, for example, by using a one dimensional signature of a compressed image. A simple absolute sum of differences is used to compare the signatures of different compressed images. Images that are similar, or of the same “form” will have small differences, compared to those that are not of the same form. Using this similarity, similar images can be grouped together. Address block location information from evaluated images in a group can be used to help determine the address block location of images that are added to the group.
REFERENCES:
patent: 3884370 (1975-05-01), Bradshaw et al.
patent: 4484348 (1984-11-01), Shizuno
patent: 4516264 (1985-05-01), Corvari et al.
patent: 4632252 (1986-12-01), Haruki et al.
patent: 4741047 (1988-04-01), Sharpe, II
patent: 4998626 (1991-03-01), Ota
patent: 5031206 (1991-07-01), Riskin
patent: 5079714 (1992-01-01), Manduley et al.
patent: 5386482 (1995-01-01), Basso et al.
Gozzo Francesco
Higgins Joseph T.
Couso Jose L.
Do Anh Hong
Lockheed Martin Corporation
McGuireWoods LLP
LandOfFree
Method for exploiting correlated mail streams using optical... 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 for exploiting correlated mail streams using optical..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for exploiting correlated mail streams using optical... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2494309