Fake signature detection

1 . A method comprising:

identifying, using one or more processors, within a document image captured using a camera, a signature portion of the document image, the signature portion representing a signature in the document image, the document image captured by a camera;

applying, using the one or more processors, a variational autoencoder to the signature portion of the document image, wherein the variational autoencoder disentangles a style of the signature portion of the document image from a content of the signature portion of the document image;

determining, using the one or more processors, whether the signature represented in the signature portion of the document image is a fake signature including:

determining, using an inter-signature detection model, whether a commonality exists between the signature represented in the document image and other signature instances, wherein the commonality is based on one or more of the content of the signature portion of the document image and the style of the signature portion of the document image; and

determining, using an intra-signature detection model, whether an anomalous consistency exists between a first character and a second character, wherein the first character and second character are characters within the signature represented in the document image; and

presenting, using the one or more processors, the determination.

2 . The method of claim 1 , wherein identifying the signature portion of the document image includes:

performing object detection to detect the signature portion of the document image.

3 . The method of claim 1 , wherein the other signature instances are associated with other document instances that include different content information.

4 . The method of claim 1 , wherein the first character and the second character are different instances of a same character used multiple times within the signature.

5 . The method of claim 1 , wherein the anomalous consistency includes one or more of anomalously consistent character height, anomalously consistent character width, and anomalously consistent character spacing.

6 . The method of claim 1 , wherein the intra-signature detection model is taught to learn the anomalous consistency using prior signatures as training data.

7 . The method of claim 1 , wherein determining that the signature is fake is based on the intra-signature detection model determining one or more characters within the signature are consistent with a computer font.

8 . The method of claim 1 , further comprising:

based on determining that the signature is fake, taking an action, wherein the action includes one or more of rejecting a user request, subjecting the user request to an additional authentication, contacting an authority, initiating a fraud case, and labeling the user request as fraudulent.

9 . The method of claim 1 , wherein the document is one of a government document, school document, financial document, and legal document.

10 . A system comprising:

one or more processors; and

a memory, the memory storing instructions that, when executed by the processor, cause the system to:

identify, within a document image captured using a camera, a signature portion, the signature portion representing a signature;

apply a variational autoencoder to the signature portion of the document image, wherein the variational autoencoder disentangles a style of the signature portion of the document image from a content of the signature portion of the document image;

determine whether the signature represented in the signature portion of the document image is a fake signature including:

determining, using an inter-signature detection model, whether a commonality exists between the signature represented in the document image and other signature instances, wherein the commonality is based on one or more of the content of the signature portion of the document image and the style of the signature portion of the document image; and

determining, using an intra-signature detection model, whether an anomalous consistency exists between a first character and a second character, wherein the first character and second character are characters within the signature represented in the document image; and

present the determination.

11 . The system of claim 10 , wherein the instructions for identifying the signature portion of the document image further cause the system to:

perform object detection to detect the signature portion.

12 . The system of claim 10 , wherein the other signature instances associated with other document instances that include different content information.

13 . The system of claim 10 , wherein the first character and the second character are different instances of a same character used multiple times within the signature.

14 . The system of claim 10 , wherein the anomalous consistency includes one or more of anomalously consistent character height, anomalously consistent character width, and anomalously consistent character spacing.

15 . The system of claim 10 , wherein the intra-signature detection model is taught to learn the anomalous consistency using prior signatures as training data.

16 . The system of claim 10 , wherein determining that the signature is fake is based on the intra-signature detection model determining one or more characters within the signature are consistent with a computer font.

17 . The system of claim 10 , wherein the instructions further cause the system to:

based on determining that the signature is fake, take an action, wherein the action includes one or more of rejecting a user request, subjecting the user request to an additional authentication, contacting an authority, initiating a fraud case, and labeling the user request as fraudulent.

18 . The system of claim 10 , wherein the document is one of a government document, school document, financial document, and legal document.