Tamper detection for identification documents

Abstract

Methods for detecting digital or physical tampering of an imaged physical credential include the actions of: receiving a digital image representing a physical credential having one or more high value regions, the digital image including an array of pixels; processing the digital image with a tamper detector to generate an output corresponding to an intrinsic characteristic of the digital image, the tamper detector configured to perform a pixel-level analysis of the high value regions of the digital image with respect to a predetermined tampering signature; and determining, based on the output from the tamper detector, whether the digital image has been digitally tampered with.

1 .- 20 . (canceled) 
     
     
         21 . A method for detecting a counterfeit credential, the method comprising:
 obtaining, using one or more computers, an image depicting a physical credential;   generating, using one or more computers, input data representing the obtained image;   providing, using one or more computers, the generated input data as an input to a machine learning model that has been trained to generate output data indicating a likelihood that an image, of a physical credential, has been digitally tampered with based on processing an image of the physical credential;   processing, using one or more computers, the provided input data through the machine learning model to generate output data indicating a likelihood that an image, of a physical credential, has been digitally tampered with based on processing an image of the physical credential;   determining, using one or more computers, whether the generated output data satisfies a predetermined threshold; and   based on a determination that the generated output data satisfies the predetermined threshold, generating, by the one or more computers, an alert indicating that the physical credential depicted by the obtained image is likely a counterfeit credential.   
     
     
         22 . The method of  claim 21 , wherein the machine learning model has been trained on a plurality of training data images, wherein the training data images comprise:
 a first plurality of images, wherein each particular image in the first plurality of images have each been labeled with a first label type, the first label type indicating that the particular image in the first plurality of images is an image, of a physical credential, that has been digitally tampered with; and   a second plurality of images, wherein each particular image in the second plurality of images have been labeled with a second label type, the second label type indicating that the particular image in the second plurality of images is an image, of a physical credential, that has not been digitally tampered with.   
     
     
         23 . The method of  claim 22 , wherein a training data image, of a physical credential, that has been digitally tampered with in the first plurality of images comprises an image depicting one or more indications of a digital modification to an image of a physical credential. 
     
     
         24 . The method of  claim 23 , wherein the digital modification comprises a digital modification to the image of the physical credential in the first plurality of images that misrepresents a feature of the imaged physical credential. 
     
     
         25 . The method of  claim 24 , wherein the digital modification to the image of the physical credential the first plurality of images that misrepresents a feature of the imaged physical credential comprises (i) a digital replacement of a photo in the image of the physical credential with a different photo, (ii) a digital replacement of biographic information in the image of the physical credential with different biographic information, (iii) a digital replacement of a machine-readable zone (MRZ) in the image of the physical credential with a different MRZ, (iv) a digital occlusion of a photo in the image of the physical credential, (v) a digital occlusion of biographic information in the image of the physical credential, (vi) a digital occlusion of an MRZ in the image of the physical credential. 
     
     
         26 . The method of  claim 25 , wherein digital occlusion can include digital blurring. 
     
     
         27 . The method of  claim 21 , wherein the method further comprising:
 based on a determination that the generated output data does not satisfy the predetermined threshold, generating output data indicating that the physical credential is likely not a counterfeit credential.   
     
     
         28 . A system for detecting a counterfeit credential, the system comprising:
 one or more computers; and   one or more memory devices storing instructions that, when executed by the one or more computers, causes the one or more computers to perform operations comprising:
 obtaining, using the one or more computers, an image depicting a physical credential; 
 generating, using the one or more computers, input data representing the obtained image; 
 providing, using the one or more computers, the generated input data as an input to a machine learning model that has been trained to generate output data indicating a likelihood that an image, of a physical credential, has been digitally tampered with based on processing an image of the physical credential; 
 processing, using the one or more computers, the provided input data through the machine learning model to generate output data indicating a likelihood that an image, of a physical credential, has been digitally tampered with based on processing an image of the physical credential; 
 determining, using the one or more computers, whether the generated output data satisfies a predetermined threshold; and 
 based on a determination that the generated output data satisfies the predetermined threshold, generating, by the one or more computers, an alert indicating that the physical credential depicted by the obtained image is likely a counterfeit credential. 
   
     
     
         29 . The system of  claim 28 , wherein the machine learning model has been trained on a plurality of training data images, wherein the training data images comprise:
 a first plurality of images, wherein each particular image in the first plurality of images have each been labeled with a first label type, the first label type indicating that the particular image in the first plurality of images is an image, of a physical credential, that has been digitally tampered with; and   a second plurality of images, wherein each particular image in the second plurality of images have been labeled with a second label type, the second label type indicating that the particular image in the second plurality of images is an image, of a physical credential, that has not been digitally tampered with.   
     
     
         30 . The system of  claim 29 , wherein a training data image, of a physical credential, that has been digitally tampered with in the first plurality of images comprises an image depicting one or more indications of a digital modification to an image of a physical credential. 
     
     
         31 . The system of  claim 30 , wherein the digital modification comprises a digital modification to the image of the physical credential in the first plurality of images that misrepresents a feature of the imaged physical credential. 
     
     
         32 . The system of  claim 31 , wherein the digital modification to the image of the physical credential the first plurality of images that misrepresents a feature of the imaged physical credential comprises (i) a digital replacement of a photo in the image of the physical credential with a different photo, (ii) a digital replacement of biographic information in the image of the physical credential with different biographic information, (iii) a digital replacement of a machine-readable zone (MRZ) in the image of the physical credential with a different MRZ, (iv) a digital occlusion of a photo in the image of the physical credential, (v) a digital occlusion of biographic information in the image of the physical credential, (vi) a digital occlusion of an MRZ in the image of the physical credential. 
     
     
         33 . The system of  claim 32 , wherein digital occlusion can include digital blurring. 
     
     
         34 . The system of  claim 28 , wherein the operations further comprising:
 based on a determination that the generated output data does not satisfy the predetermined threshold, generating output data indicating that the physical credential is likely not a counterfeit credential.   
     
     
         35 . A computer-readable storage medium storing instructions that, when executed by one or more computers, causes the one or more computers to perform operations for detecting a counterfeit credential, the operations comprising:
 obtaining an image depicting a physical credential;   generating input data representing the obtained image;   providing the generated input data as an input to a machine learning model that has been trained to generate output data indicating a likelihood that an image, of a physical credential, has been digitally tampered with based on processing an image of the physical credential;   processing the provided input data through the machine learning model to generate output data indicating a likelihood that an image, of a physical credential, has been digitally tampered with based on processing an image of the physical credential;   determining whether the generated output data satisfies a predetermined threshold; and   based on a determination that the generated output data satisfies the predetermined threshold, generating an alert indicating that the physical credential depicted by the obtained image is likely a counterfeit credential.   
     
     
         36 . The computer-readable storage medium of  claim 35 , wherein the machine learning model has been trained on a plurality of training data images, wherein the training data images comprise:
 a first plurality of images, wherein each particular image in the first plurality of images have each been labeled with a first label type, the first label type indicating that the particular image in the first plurality of images is an image, of a physical credential, that has been digitally tampered with; and   a second plurality of images, wherein each particular image in the second plurality of images have been labeled with a second label type, the second label type indicating that the particular image in the second plurality of images is an image, of a physical credential, that has not been digitally tampered with.   
     
     
         37 . The computer-readable storage medium of  claim 36 , wherein a training data image, of a physical credential, that has been digitally tampered with in the first plurality of images comprises an image depicting one or more indications of a digital modification to an image of a physical credential. 
     
     
         38 . The computer-readable storage medium of  claim 37 , wherein the digital modification comprises a digital modification to the image of the physical credential in the first plurality of images that misrepresents a feature of the imaged physical credential. 
     
     
         39 . The computer-readable storage medium of  claim 38 , wherein the digital modification to the image of the physical credential the first plurality of images that misrepresents a feature of the imaged physical credential comprises (i) a digital replacement of a photo in the image of the physical credential with a different photo, (ii) a digital replacement of biographic information in the image of the physical credential with different biographic information, (iii) a digital replacement of a machine-readable zone (MRZ) in the image of the physical credential with a different MRZ, (iv) a digital occlusion of a photo in the image of the physical credential, (v) a digital occlusion of biographic information in the image of the physical credential, (vi) a digital occlusion of an MRZ in the image of the physical credential. 
     
     
         40 . The computer-readable storage medium of  claim 35 , wherein the operations further comprising:
 based on a determination that the generated output data does not satisfy the predetermined threshold, generating output data indicating that the physical credential is likely not a counterfeit credential.