Automatic system and method for document authentication using portrait fraud detection
An authentication processing system includes a memory storing a portrait fraud detection application, and a processing unit coupled with the memory and configured to execute the portrait fraud detection application. The portrait fraud detection application, when executed, configures the processing unit to receive a capture of a document including a portrait photo and at least one overlay, detect a face in the portrait photo among the at least one overlay in the capture, and determine the portrait photo is fraudulent; and initiate an indication the document is fraudulent.
1. An authentication processing system, comprising:
a memory storing a portrait fraud detection application; and
a processing unit coupled with the memory and configured to execute the portrait fraud detection application, the portrait fraud detection application, when executed, configuring the processing unit to:
receive a capture of a document including a portrait photo and at least one overlay;
detect a face within the portrait photo among the at least one overlay in the capture;
determine the portrait photo is fraudulent, wherein the processing unit determines the photo is fraudulent by:
rendering-flat the document;
computing edges of the portrait photo within the document;
computing candidate boundary lines from the edges;
computing a portrait frame from the candidate boundary lines; and
computing a fake boundary confidence value for the portrait frame, the fake boundary confidence value exceeding a threshold to determine the portrait photo is fraudulent; and
initiate an indication the document is fraudulent.
2. The authentication processing system of claim 1 , wherein the processing unit detects the face within portrait photo using a single shot detector (SSD) algorithm.
3. The authentication processing system of claim 2 , wherein the processing unit is further configured to train the SSD algorithm using sample portrait photos having overlays.
4. The authentication processing system of claim 3 , wherein the processing unit determines the portrait photo is fraudulent further by:
executing the SSD algorithm to detect a ghost photo in the document;
applying masks to the one or more overlays in the portrait photo and the ghost photo;
executing a template matching algorithm to match the portrait photo and the ghost photo; and
detecting the ghost photo does not match the portrait photo.
5. The authentication processing system of claim 1 , wherein computing edges of the portrait photo comprises computing local-type edges and rim-type edges.
6. The authentication processing system of claim 1 , wherein the processing unit determines the portrait photo is fraudulent further by:
computing a first portrait profile template for a document type corresponding to the document received;
computing a second portrait profile template for the document;
comparing the first portrait profile template to the second portrait profile template; and
determining the portrait photo is fraudulent when the first portrait profile template does not match the second portrait profile.
7. The authentication processing system of claim 1 , wherein the memory and the processing unit are components of a mobile device.
8. The authentication processing system of claim 1 , wherein the at least one overlay comprises an overlay selected from the group consisting of:
text,
a hologram, and
a security pattern.
9. A method of detecting a fraudulent portrait photo in a document, the method comprising:
receiving a capture of a document including a portrait photo;
detecting a face within the portrait photo in the capture;
determining the portrait photo is fraudulent, wherein determining the portrait photo is fraudulent comprises:
rendering-flat the document;
computing edges of the portrait photo within the document;
computing candidate boundary lines from the edges;
computing a portrait frame from the candidate boundary lines; and
computing a fake boundary confidence value for the portrait frame, the fake boundary confidence value exceeding a threshold to determine the portrait photo is fraudulent; and
initiating an indication the document is fraudulent.
10. The method of claim 9 , wherein detecting the face within the portrait photo comprises using a single shot detector (SSD) algorithm.
11. The method of claim 10 , wherein detecting the face within the portrait photo further comprises training the SSD algorithm using sample portrait photos having overlays.
12. The method of claim 11 , wherein determining the portrait photo is fraudulent further comprises:
executing the SSD algorithm to detect a ghost photo in the document;
applying masks to overlays in the portrait photo and the ghost photo;
executing a template matching algorithm to match the portrait photo and the ghost photo; and
detecting the ghost photo does not match the portrait photo.
13. The method of claim 9 , wherein computing edges of the portrait photo comprises computing local-type edges and rim-type edges.
14. The method of claim 9 , wherein determining the portrait photo is fraudulent further comprises:
computing a first portrait profile template for a document type corresponding to the document received;
computing a second portrait profile template for the document;
comparing the first portrait profile template to the second portrait profile template; and
determining the portrait photo is fraudulent when the first portrait profile template does not match the second portrait profile.
15. A method of detecting a fraudulent portrait photo boundary in a document, the method comprising:
rendering-flat the document;
computing edges of a portrait photo within the document;
computing candidate boundary lines from the edges;
computing a portrait frame from the candidate boundary lines; and
computing a fake boundary confidence value for the portrait frame, the fake boundary confidence value exceeding a threshold to determine the portrait photo is fraudulent.
16. The method of claim 15 , wherein computing edges of the portrait photo comprises computing local-type edges and rim-type edges.
17. The method of claim 15 , wherein computing candidate boundary lines comprises applying a Hough transform to the edges to identify true edges of the portrait photo.
18. The method of claim 15 , wherein computing the portrait frame comprises:
computing a portrait center based on a region of interest for the portrait photo;
categorizing the candidate boundary lines as top, bottom, left, or right based on the portrait center;
merging the candidate boundary lines within each category;
selecting the best boundary line for each category based on peakedness; and
computing frame corners based on computed intersection points of two or more of the best boundary lines.
19. The method of claim 15 , wherein computing a fake boundary confidence value comprises computing a value based on angles computed between adjacent frame boundary lines.
20. The method of claim 15 , wherein computing a fake boundary confidence value comprises computing a value based on a ratio of a segment length from the portrait frame to a corresponding portrait region of interest dimension.