Method and system for detecting anomalies in a portable document format (PDF) document

1 . A method for analyzing a portable document format (PDF) document, the method comprising:

modifying, by a PDF analyzer, a background color of the PDF document to a second color to generate a modified PDF document, wherein the background color is determined by analyzing a color of the PDF document at an outer margin of the PDF document that is outside of a printable area and wherein the background color of the PDF document is modified by changing a rendering instruction associated with the PDF document;

converting, by a noise removal module (NRM), the modified PDF document into an image, wherein the modified PDF document is sent by the PDF analyzer and the image includes multiple pixels that are arranged in a two dimensional map;

making, by the NRM, a first determination that the image comprises a background noise, wherein the background noise is reduced by modifying a background-colored pixel of an area within the image;

generating, by a fraud detection module (FDM), at least one image patch from the image, wherein the image is sent by the NRM,

wherein generating the at least one image patch comprises:

processing the image to identify a plurality of regions of interest, wherein each of the plurality of regions of interest comprises a pixel;

generating a plurality of image patches, wherein each of the image patches is centered around one of the plurality of pixels, wherein the at least one image patch is one of the plurality of image patches;

generating, by the FDM, an embedding for the at least one image patch, wherein the embedding is processed using a convolutional neural network (CNN) layer to classify the PDF document;

making, by the FDM and based on an output of the CNN layer, a second determination that the PDF document is valid;

extracting, by the FDM and based on the second determination, metadata associated with the PDF document;

making, by the FDM and based on the metadata, a third determination that the PDF document is invalid, wherein the metadata specifies that the PDF document is generated using an invalid PDF document generator based on a list of trusted PDF document generators in a database; and

notifying, by the FDM, a user of a vendor environment (VE) about the third determination using a graphical user interface (GUI) of the VE.

2 . The method of claim 1 , wherein the background noise indicates a size of the area has exceeded a predetermined noise threshold.

3 . The method of claim 1 , wherein the CNN layer comprises at least one selected from a group consisting of a Fourier transform layer, a normalization layer, a softmax layer, and a residual connection.

4 . The method of claim 1 , wherein the embedding is a low-dimensional vector representation of the at least one image patch, wherein the FDM employs a linear projection model to generate the embedding.

5 . The method of claim 4 , wherein the low-dimensional vector representation of the at least one image patch reduces computation complexity that needs to be managed by the FDM.

6 . The method of claim 1 , wherein the background color of the PDF document is modified by changing a rendering instruction associated with the PDF document.

7 . The method of claim 1 , wherein the third determination is made by employing a Levenshtein distance calculation model.

8 . A method for analyzing a portable document format (PDF) document, the method comprising:

modifying, by a PDF analyzer, a background color of the PDF document to a second color to generate a modified PDF document, wherein the background color is determined by analyzing a color of the PDF document at an outer margin of the PDF document that is outside of a printable area and wherein the background color of the PDF document is modified by changing a rendering instruction associated with the PDF document;

converting, by a noise removal module (NRM), the modified PDF document into an image, wherein the modified PDF document is sent by the PDF analyzer and the image includes multiple pixels that are arranged in a two dimensional map;

making, by the NRM, a first determination that the image comprises a background noise, wherein the background noise is reduced by modifying a background-colored pixel of an area within the image;

generating, by a fraud detection module (FDM), at least one image patch from the image, wherein the image is sent by the NRM,

wherein generating the at least one image patch comprises:

processing the image to identify a plurality of regions of interest, wherein each of the plurality of regions of interest comprises a pixel;

generating a plurality of image patches, wherein each of the image patches is centered around one of the plurality of pixels, wherein the at least one image patch is one of the plurality of image patches;

generating, by the FDM, an embedding for the at least one image patch, wherein the embedding is processed using a convolutional neural network (CNN) layer to classify the PDF document;

making, by the FDM and based on an output of the CNN layer, a second determination that the PDF document is valid;

extracting, by the FDM and based on the second determination, metadata associated with the PDF document;

making, by the FDM and based on the metadata, a third determination that the PDF document is invalid, wherein the metadata specifies that the PDF document is generated using a valid PDF document generator based on a list of trusted PDF document generators in a database; and

notifying, by the FDM, a user of a vendor environment (VE) about the second determination and the third determination using a graphical user interface (GUI) of the VE.

9 . The method of claim 8 , wherein the background noise indicates a size of the area has exceeded a predetermined noise threshold.

10 . The method of claim 8 , wherein the CNN layer comprises at least one selected from a group consisting of a Fourier transform layer, a normalization layer, a softmax layer, and a residual connection.

11 . The method of claim 8 , wherein the embedding is a low-dimensional vector representation of the at least one image patch, wherein the FDM employs a linear projection model to generate the embedding.

12 . The method of claim 11 , wherein the low-dimensional vector representation of the at least one image patch reduces computation complexity that needs to be managed by the FDM.

13 . The method of claim 8 , wherein the background color of the PDF document is modified by changing a rendering instruction associated with the PDF document.

14 . A method for analyzing a portable document format (PDF) document, the method comprising:

modifying, by a PDF analyzer, a background color of the PDF document to a second color to generate a modified PDF document, wherein the background color is determined by analyzing a color of the PDF document at an outer margin of the PDF document that is outside of a printable area and wherein the background color of the PDF document is modified by changing a rendering instruction associated with the PDF document;

converting, by a noise removal module (NRM), the modified PDF document into an image, wherein the modified PDF document is sent by the PDF analyzer and the image includes multiple pixels that are arranged in a two dimensional map;

making, by the NRM, a first determination that the image comprises a background noise, wherein the background noise is reduced by modifying a background-colored pixel of an area within the image;

generating, by a fraud detection module (FDM), at least one image patch from the image, wherein the image is sent by the NRM,

wherein generating the at least one image patch comprises:

processing the image to identify a plurality of regions of interest, wherein each of the plurality of regions of interest comprises a pixel;

generating a plurality of image patches, wherein each of the image patches is centered around one of the plurality of pixels, wherein the at least one image patch is one of the plurality of image patches;

generating, by the FDM, an embedding for the at least one image patch, wherein the embedding is processed using a convolutional neural network (CNN) layer to classify the PDF document;

making, by the FDM and based on an output of the CNN layer, a second determination that the PDF document is invalid; and

notifying, by the FDM, a user of a vendor environment (VE) about the second determination using a graphical user interface (GUI) of the VE.

15 . The method of claim 14 , wherein the background noise indicates a size of the area has exceeded a predetermined noise threshold.

16 . The method of claim 14 , wherein the CNN layer comprises at least one selected from a group consisting of a Fourier transform layer, a normalization layer, a softmax layer, and a residual connection.

17 . The method of claim 14 , wherein the embedding is a low-dimensional vector representation of the at least one image patch, wherein the FDM employs a linear projection model to generate the embedding.

18 . The method of claim 17 , wherein the low-dimensional vector representation of the at least one image patch reduces computation complexity that needs to be managed by the FDM.

19 . The method of claim 14 , wherein the background color of the PDF document is modified by changing a rendering instruction associated with the PDF document.