Check image data interference processing
View Patent ↗Various embodiments herein each include at least one of systems, methods, and software for check image data inference processing. Another example method embodiment includes inferring a check amount of a check image included in an account group of check images stored in a memory device. Where the check amount is missing in check data associated with the check image or was poorly read by an optical character recognition process, the method includes inferring of the check amount based at least in part on one or more check amounts of check data associated with other check images of the account group. Once inferred, the method includes updating the check amount of the check data associated with the respective check image with the inferred check amount of the check image. Some embodiments also or alternatively include inferring a check date.
1. A method comprising:
extracting, by executing instructions on a computer processor, at least a portion of text included in each digital document image of a plurality of check images and storing, on a data storage device, the extracted text as check data associated with a text field of the respective check image;
identifying a plurality of checks in the stored check data that are associated with a same account;
identifying a text field in the stored data of one check from which text was not reliably extracted; and
inferring, by executing instructions on the computer processor, content of the text field based on the extracted text and on text extracted from corresponding text fields in check data of other check images associated with the same account, the inferring including a distance comparison according to a distance measuring algorithm that measures a distance between extracted text and text extracted from corresponding text fields in check data of other check images associated with the same account; and
storing the inferred content on the data storage device.
2. The method of claim 1 , wherein the check images are received with check code line data read from a code line of each respective check, the check code line data included in check data associated with a respective check image.
3. The method of claim 2 , wherein the extracting includes performing optical character recognition to extract a portion of text from a text field and adding the text to the check data associated with a respective check image from which the text was extracted.
4. The method of claim 3 , wherein:
performing the optical character recognition further obtains a confidence value with regard to the extracted text, a confidence value obtained with specific regard to extracted text of at least one of a date and an amount of a check image, the confidence value added to check data associated with the respective check image; and
the inferring of content of the text field from which text was not reliability extracted is performed when a confidence value of a date or amount read from a check image is below a confidence threshold.
5. The method of claim 4 , wherein when inferring date content for a text field, the inferring of the date content includes:
ordering the plurality of documents associated with the same account as the document from which the text field was not reliability extracted by at least one of dates and check numbers extracted from the respective documents;
identifying a frequency of dates of the check images; and
inferring the date content of the text field that was not reliability extracted based at least in part on the identified frequency of check image dates.
6. The method of claim 4 , wherein a text field that was not reliably extracted includes an ambiguously read date, an ambiguously read date including a date extracted from a check image where a month and day are both represented as a number less than thirteen (13), the inferring of the text content of an ambiguously read date performed according to a set of ambiguity resolution rules, the ambiguity resolution rules including:
identifying an order of month and day extracted from at least one other check image of the documents associated with the same account and applying that order to the ambiguously read date as the inferred content;
when the ambiguously read date, when considering month/day or day/month order is a date prior to a current date, applying an order that is a future date as the inferred content; and
when one of the month/day or day/month order fills a date frequency gap between dates of other check images associated with the same account, applying an order that fills the date frequency gap as the inferred content.
7. The method of claim 4 , further comprising:
ordering the plurality of documents associated with the same account as the document from which the text field was not reliability extracted by at least one of dates and check numbers extracted from the respective documents;
when check data of a check image includes inferring a check amount that was not reliably extracted:
inferring content of the check amount of the check image based on check values included in check data of at least one other check image associated with the same account as the check image from which the check amount was not reliably extracted; and
updating the check data check amount of the check image from which the check amount was not reliably extracted.
8. The method of claim 7 , wherein inferring the check amount based on check values included in check data of at least one other check image associated with the same account as the check image from which the check amount was not reliably extracted includes:
ordering check data of check images of the plurality of documents associated with the same account by at least one of check numbers and dates;
for check amounts in check data of the plurality of documents associated with a same account having a confidence level above a confidence threshold, identifying a most common check amount;
performing a similarity comparison between the identified most common amount and the check amount extracted from the check image with the check amount that was not reliably extracted; and
when the similarity is within an acceptable tolerance, modifying the check amount of the check data of the check image that was not reliably extracted to be the same as the identified most common check amount.
9. The method of claim 8 , wherein the similarity comparison is performed according to a Levenshtein distance algorithm and the acceptable tolerance is a threshold value within which an output of the Levenshtein distance algorithm must fit to infer the check amount.
10. A method comprising:
inferring, by executing instructions on a computer processor, a check amount of a check image included in an account group of check images stored in a memory device where the check amount is missing in check data associated with the check image or was poorly read by an optical character recognition process, the inferring of the check amount based at least in part on one or more check amounts of check data associated with other check images of the account group, each stored check image including check data of text fields of the respective check images, the inferring the check amount of the check image missing or having a poorly read check amount based at least in part on one or more check amounts of check data associated with other check images of the account group includes:
ordering check data of check images within the account group by at least one of check numbers and dates;
for check amounts in check data of the account group read by an optical character recognition process with a confidence level above a confidence threshold, identifying a most common check amount;
performing a similarity comparison between the identified most common amount and the check amount of the check image missing or having a poorly read check amount, the similarity comparison including applying a distance measuring algorithm that measures a distance between extracted text and text extracted from corresponding text fields in check data of other check images associated with the same account; and
when the similarity is within an acceptable tolerance, modifying the check amount of the check data of the check image missing or having a poorly read check amount to be the same as the identified most common check amount; and
updating the check amount of the check data associated with the respective check image with the inferred check amount of the check image.
11. The method of claim 10 , wherein the similarity comparison is performed according to a Levenshtein distance algorithm and the acceptable tolerance is a threshold value within which an output of the Levenshtein distance algorithm must fit to infer the check amount.
12. The method of claim 10 , further comprising:
inferring a check date of the check image included in the account group of check images stored in the memory device where the check date is missing in the check data associated with the check image or was poorly read by the optical character recognition process, the inferring of the check date based at least in part on the check data associated with the other check images of the account group; and
updating the check date of the check data associated with the respective check image with the inferred date of the check image.
13. The method of claim 10 , further comprising:
receiving images of a plurality of post-dated checks;
performing optical character recognition, by the optical character recognition process, on at least a portion of text included in each check image to obtain the check data for each check image;
adding the check data to data associated with each respective check image; and
associating images of checks written from the same account into an account group.
14. The method of claim 13 , wherein the images of the plurality of post-dated checks are received via a network interface device.
15. A system comprising:
at least one processor;
at least one memory; and
an instruction set accessible in the memory and executable by the at least one processor, the instruction set including a set of modules, the set of modules comprising:
a check amount inference module including instructions executable by the at least one processor to perform data processing activities comprising:
inferring a check amount of a check image included in an account group of check images stored in a memory device where the check amount is missing in check data associated with the check image or was poorly read by an optical character recognition process, the inferring of the check amount based at least in part on one or more check amounts of check data associated with other check images of the account group, each stored check image including check data of text fields of the respective check images; and
updating the check amount of the check data associated with the respective check image with the inferred check amount of the check image; and
a check date inference module including instructions executable by the at least one processor to perform data processing activities comprising:
inferring a check date of the check image included in the account group of check images stored in the memory device where the check date is missing in the check data associated with the check image or was poorly read by the optical character recognition process, the inferring of the check date based on the extracted text and at least in part on the check data associated with the other check images of the account group, the inferring including a distance comparison according to a distance measuring algorithm that measures a distance between extracted text and text extracted from corresponding text fields in check data of other check images associated with the same account; and
updating the check date of the check data associated with the respective check image with the inferred date of the check image.
16. The system of claim 15 , the set of modules further comprising:
a check image receiving module including instructions executable by the at least one processor to perform data processing activities comprising:
receiving images of a plurality of checks;
performing optical character recognition, by the optical character recognition process, on at least a portion of text included in each check image to obtain the check data for each check image;
adding the check data to data associated with each respective check image;
associating images of checks written from the same account into an account group; and
ordering check data of check images within each account group by at least one of check numbers and dates.
17. The system of claim 16 , further comprising:
a network interface device; and
wherein the check image receiving modules receives the image of the plurality of checks via the network interface device.