| CPC G06Q 20/4016 (2013.01) [G06Q 20/042 (2013.01); G06V 30/19013 (2022.01); G06V 30/416 (2022.01); G06V 30/42 (2022.01)] | 20 Claims |
|
1. A method for detecting and remediating anomalies with forged institutional financial instruments, the method comprising:
obtaining a check image of a forged institutional financial instrument that appears on its face to be genuine, wherein the forged institutional financial instrument comprises one or both of:
altered information to cause the check to convey a larger amount of money than originally intended amount; and
altered information designed to cause a financial institution to allow a malicious actor to receive funds associated with the financial instrument rather than an originally intended recipient;
performing a text extraction process on the check image of the forged institutional financial instrument to obtain check textual information, wherein performing the text extraction process includes performing a template matching process using alphanumeric characters and symbols in a predetermined font, wherein the template matching process includes:
for each template of the list of templates representative of alphanumeric characters and symbols in the predetermined font, generating a match score array corresponding to each of a plurality of patches within the check image, the plurality of patches being sized corresponding to an expected character size; and
performing a cluster analysis on the match score array associated with each template to identify locations and distinct instances of each character appearing in the check image, wherein performing the cluster analysis includes clustering localized match scores above a threshold and identifying a best-match position for a character based on a highest match score within a cluster;
identifying an intelligent mail barcode (IMB) in the check image;
decoding the IMB to obtain a character code, wherein the decoding includes:
performing one or more image enhancement processes on at least a portion of the check image including the IMB, the one or more image enhancement processes including:
a tilt fixing process; and
a noise removal process, wherein the noise removal process includes:
calculating pixel densities of a track region, an ascending region, and a descending region for each of the plurality of vertical bars;
comparing the pixel densities of each of the plurality of vertical bars in the track region, the ascending region, and the descending region to detect one or more regions of one or more of the plurality of vertical bars having a lower comparative pixel density relative to corresponding regions of others of the plurality of vertical bars; and
based on detecting the one or more regions having a lower comparative pixel density, considering pixel density in the one or more regions as noise;
applying a horizontal dilation of each of a plurality of vertical bars depicted within the portion of the check image including the IMB to generate a dilated check image portion; and
applying a mask to the dilated check image portion to obtain a masked check image portion;
extracting encoded check information from the character code, including a payee ZIP code from the IMB;
performing a fraud detection process on the check image based on both:
(1) scores generated during the text extraction process, wherein performing fraud detection based on scores generated during the text extraction process includes:
determining a highest match score for the check image;
comparing the highest match score to a detection threshold; and
based on the highest match score being below the detection threshold, routing the check image for fraud analysis; and
(2) inconsistency between the check textual information and the encoded check information, wherein fraud detection based on the inconsistency includes:
comparing a ZIP code template representative of the payee ZIP code obtained from the IMB to a region of the check image containing a payee name and address to identify a presence of the payee ZIP code in the region, wherein a matching threshold used to identify the presence of the payee ZIP code is lower than a character matching threshold used in the template matching process; and
based on the fraud detection process, routing the image of the forged institutional financial instrument for additional fraud analysis and remediation rather than permitting further processing without additional fraud analysis.
|