Abstract: A method of categorizing text entries on a document can include determining, for each of a plurality of text bounding boxes in the document, respective text, respective coordinates, and respective input embeddings. The method may further include defining a graph of the plurality of bounding boxes, the graph comprising a plurality of connections among the plurality of bounding boxes, each connection comprising a first and second bounding box and zero or more respective intermediate bounding boxes. The method may further include determining a respective attention value for each connection according to a quantity of intermediate bounding boxes in the connection and, based on a the respective attention values and a transformer-based machine learning model applied to the respective input embeddings and respective coordinates, determining output embeddings for each bounding box and, based on the respective output embeddings, generating a bounding box label for each bounding box.

Abstract: Methods and systems are presented for signed document image analysis and fraud detection. An image of a document may be received from a user's device. A first layer of a machine learning engine is used to identify a signature and a name of the user within different areas of the received image. A second layer of the machine learning engine is used to extract a plurality of features from the different areas of the image. The plurality of features includes at least one visual feature representing the signature and at least one textual feature representing the name. A combined feature representation of the signature and the name is generated based on the plurality of features extracted from the image. A third layer of the machine learning engine is used to determine whether the signature of the user has been digitally altered, based on the combined feature representation.

Abstract: Methods and systems are presented for detecting physical tampering on a document based on analyzing an image of the document. When the image of the document is obtained, multiple contours are identified in the image based on pixel characteristics of the image. Dimension attributes of the contours are determined. Contours that are determined to correspond to borders or texts of the documents based on the dimension attributes are eliminated. A second text detection process based on a polygon method is performed on at least one remaining contour to determine whether the at least one remaining contour links multiple text elements together. The document is determined to have been physically manipulated when at least on contour remains in the image.

Abstract: Systems and/or techniques for facilitating image forgery detection via pixel-metadata consistency analysis are provided. In various embodiments, a system can receive an electronic image from a client device. In various cases, the system can obtain a pixel vector and/or an image metadata vector that correspond to the electronic image. In various aspects, the system can determine whether the electronic image is authentic or forged, based on analyzing the pixel vector and the image metadata vector via at least one machine learning model.

Abstract: Systems and/or techniques for facilitating image forgery detection via headpose estimation are provided. In various embodiments, a system can receive a document from a client device. In various cases, the system can identify, by executing a first trained machine learning model, an object that is depicted in the document. In various instances, the system can determine, by executing a second trained machine learning model, a pose of the object. In various aspects, the system can determine, by executing a third trained machine learning model, whether the document is authentic or forged based on the pose of the object. In various embodiments, the system can, in response to determining that the document is forged, transmit an unsuccessful validation message to the client device.