Method and system for verifying uploaded customer photographs

A method for automatically validating an image uploaded to a computer system by a customer, the system comprising:

• generating an upload prompt to request a customer to upload an image of a requested type to a computer system, wherein the requested type of image comprises one of a headshot of the customer, a front of an identification card of the customer, or a back of an identification card of the customer;
• receiving with the computer system an image uploaded in response to the upload prompt;
• automatically processing the uploaded image with a classification model operative to return (i) a first probability that the uploaded image comprises an image of a headshot, (ii) a second probability that the uploaded image comprises an image of a front of an identification card, and (iii) a third probability that the uploaded image comprises an image of a back of an identification card;
• using the returned first probability, second probability and third probability to classify the image as one of a headshot, a front of a driver's license, a back of a driver's license, or a rogue image;
• if the classification model classifies the uploaded image as a headshot, automatically processing the uploaded image with a face detection model operative to return a headshot return result that includes a face bounding box for the uploaded image;
• if the classification model classifies the uploaded image as a front of an identification card:
• automatically processing the uploaded image with an identification card detection model to return an identification card front return result that includes an identification card front bounding box for the uploaded image;
• and
• using the identification card front bounding box to process the uploaded image with the face detection model to return an identification card headshot return result that includes an identification card face bounding box for the uploaded image;
• and
• if the classification model classifies the uploaded image as a back of an identification card, processing the uploaded image with the identification card detection model to return an identification card back return result that includes an identification card back bounding box for the uploaded image;
• wherein the classification model is implemented via a convolutional neural network having an architecture that comprises a modification of an Inception v3 architecture, wherein the modification of the Inception v3 architecture includes a global max pool layer, a fully connected layer and an output max layer in place of a fully connected layer of the Inception v3 architecture.

A computerized image verification system for automatically verifying an image uploaded by a customer, the image verification system comprising:

• an input component operative to receive images uploaded to the system;
• a storage component operative to store information relating to the images uploaded to the system;
• and
• a processor in communication with the input component and the storage component and operative:
• to receive an image uploaded to the customer verification system in response to an upload prompt requesting a customer to upload an image having an image type of a headshot of the customer, a front of an identification card of the customer, or a back of an identification card of the customer;
• to automatically input the uploaded image to a classification neural network trained to attempt to classify the type of the uploaded image as one of a headshot, a front of an identification card, or a back of an identification card;
• if the classification neural network classifies the uploaded image as a headshot, to automatically input the uploaded image to a face detection model configured to return a headshot return result that includes a face bounding box for the uploaded image;
• if the classification neural network classifies the uploaded image as a front of an identification card:
• to automatically input the uploaded image to an identification card detection neural network operative to return an identification card return result that includes an identification card front bounding box for the uploaded image;
• and
• to use the identification card front bounding box to process the uploaded image with the face detection model to return an identification card headshot return result that includes a face bounding box of a detected face for the identification card front bounding box for the uploaded image;
• and
• if the classification neural network classifies the uploaded image as a back of an identification card, processing the uploaded image with the identification card detection neural network to return a return result that includes an identification card back bounding box for the uploaded image;
• wherein the classification neural network comprises a convolutional neural network having an architecture that comprises a modification of an Inception v3 architecture, wherein the modification includes a global max pool layer, a fully connected layer and an output max layer in place of a fully connected layer of the Inception v3 architecture.

A method for automatically verifying an image uploaded to a computer system by a customer, the method comprising:

• generating an upload prompt to request a customer to upload an image of a requested type to a computer system, wherein the requested type of image comprises one of a headshot of the customer, a front of an identification card of the customer, or a back of an identification card of the customer;
• receiving with the computer system an image uploaded in response to the upload prompt;
• automatically processing the uploaded image with a computer-implemented classification model operative to attempt to classify the uploaded image as one of a headshot, a front of an identification card, or a back of an identification card;
• and
• comparing the classification of the uploaded image with the image type requested by the upload prompt to verify that the image type of the uploaded image matches the image type requested by the upload prompt;
• wherein the classification model is implemented via a convolutional neural network having an architecture that comprises a modification of an Inception v3 architecture, wherein the modification of the Inception v3 architecture includes a global max pool layer, a fully connected layer and an output max layer in place of a fully connected layer of the Inception v3 architecture.