System and method for detecting signature forgeries

Abstract

Two models are first trained and then test images are applied to the two trained models in an effort to detect signature forgeries. The first model is trained with pairs of signature images and the resultant trained model is capable of detecting blind forgeries. The second model is trained with triplets of signature images and is capable of detecting skilled signature forgeries. After the two models are trained, test images are applied to the models and determinations are made as to whether a blind or skilled forgery is present.

What is claimed is: 
     
         1 . A system for detecting forged information in a retail environment, the system comprising:
 a first image capture device that is configured to obtain first images of signatures;   a second image capture device;   an electronic memory device that stores a first mathematical model and a second mathematical model;   a control circuit, the control circuit being configured to:   train the first mathematical model with a first set of the first images obtained from the first image capture device, the training being effective to allow the first mathematical model to detect blind signature forgeries and to produce a trained first mathematical model, the training of the first mathematical model resulting in a physical transformation of the first mathematical model;   train the second mathematical model with a second set of the first images obtained from the first image capture device, the training of the second mathematical model being effective to allow the second mathematical model to detect skilled signature forgeries and to produce a trained second mathematical model, the training of the second mathematical model resulting in a physical transformation of the second mathematical model;   subsequently, receive a test image from the second image capture device, the test image including a second signature that is associated with a transaction concerning a retail environment;   apply the test image to the trained first mathematical model, and when an application determines the second signature in the test image is a blind signature forgery, perform a first action;   when the application does not detect a blind signature forgery, apply the test image to the trained second mathematical model to determine whether the second signature in the test image is a skilled forgery;   when the second signature in the test image is determined to be a skilled forgery, perform a second action;   when the second signature in the test image is determined not to be a skilled forgery, perform a third action;   wherein the first action, the second action, and the third action include one or more of: halting a retail transaction, finalizing or validating a retail transaction, issuing an electronic alert, issuing an electronic inquiry, accepting merchandise from a supplier at a loading dock, or releasing merchandise to a customer using an automated vehicle or robot.   
     
     
         2 . The system of  claim 1 , wherein the training of the first mathematical model includes determining a dissimilarity score between the pairs of the first images. 
     
     
         3 . The system of  claim 1 , wherein Bayesian optimization is utilized to optimize parameters of the first mathematical model and the second Mathematical model. 
     
     
         4 . The system of  claim 1 , wherein the first set of images comprises triplets and each triplet comprises an anchor image, a positive sample image, and a negative sample image. 
     
     
         5 . The system of  claim 1 , wherein the first mathematical model and the second mathematical model are convolutional neural networks. 
     
     
         6 . The system of  claim 1 , wherein the control circuit comprises a first control circuit and a second control circuit, the first control circuit being disposed at a central location remote from the retail store, the second control circuit being associated with a mobile electronic device running software application and being disposed at the retail store. 
     
     
         7 . The system of  claim 1 , wherein the signatures are associated with retail customers or business suppliers. 
     
     
         8 . The system of  claim 1 , wherein the signatures are disposed on paper documents or are electronic signatures. 
     
     
         9 . The system of  claim 1 , wherein the second image capture device is deployed at a retail store. 
     
     
         10 . The system of  claim 1 , wherein the electronic alert is sent to an employee of the retail store. 
     
     
         11 . A method for detecting forged information in a retail environment, the method comprising:
 configuring a first image capture device to obtain first images of signatures;   deploying a second image capture device;   storing a first mathematical model and a second mathematical model in an electronic memory device;   by a control circuit, training the first mathematical model with a first set of the first images obtained from the first image capture device, the training being effective to allow the first mathematical model to detect blind signature forgeries and to produce a trained first mathematical model, the training of the first mathematical model resulting in a physical transformation of the first mathematical model;   by the control circuit, training the second mathematical model with a second set of the first images obtained from the first image capture device, the training of the second mathematical model being effective to allow the second mathematical model to detect skilled signature forgeries and to produce a trained second mathematical model, the training of the second mathematical model resulting in a physical transformation of the second mathematical model;   subsequently, receiving at the control circuit a test image from the second image capture device, the test image including a second signature that is associated with a transaction concerning a retail environment;   by the control circuit, applying the test image to the trained first mathematical model, and when an application determines the second signature in the test image is a blind signature forgery, performing a first action;   by the control circuit and when the application does not detect a blind signature forgery, applying the test image to the trained second mathematical model to determine whether the second signature in the test image is a skilled forgery;   by the control circuit and when the second signature in the test image is determined to be a skilled forgery, performing a second action;   by the control circuit and when the second signature in the test image is determined not to be a skilled forgery, performing a third action;   wherein the first action, the second action, and the third action include one or more of: halting a retail transaction, finalizing or validating a retail transaction, issuing an electronic alert, issuing an electronic inquiry, accepting merchandise from a supplier at a loading dock, or releasing merchandise to a customer using an automated vehicle or robot.   
     
     
         12 . The method of  claim 11 , wherein the training of the first mathematical model includes determining a dissimilarity score between the pairs of the first images. 
     
     
         13 . The method of  claim 11 , further comprising using Bayesian optimization to optimize parameters of the first mathematical model and the second mathematical model. 
     
     
         14 . The method of  claim 11 , wherein the first set of images comprises triplets and each triplet comprises an anchor image, a positive sample image, and a negative sample image. 
     
     
         15 . The method of  claim 11 , wherein the first mathematical model and the second mathematical model are convolutional neural networks. 
     
     
         16 . The method of  claim 11 , wherein the control circuit comprises a first control circuit and a second control circuit, the first control circuit being disposed at a central location remote from the retail store, the second control circuit being associated with a mobile electronic device running software application and being disposed at the retail store. 
     
     
         17 . The method of  claim 11 , wherein the signatures are associated with retail customers or business suppliers. 
     
     
         18 . The method of  claim 11 , wherein the signatures are disposed on paper documents or are electronic signatures. 
     
     
         19 . The method of  claim 11 , wherein the second image capture device is deployed at a retail store. 
     
     
         20 . The method of  claim 11 , wherein the electronic alert is sent to an employee of the retail store.