Distributed computer system for document authentication

1. A method comprising:

generating an image of a document to be audited;

pre-processing the image to obtain at least one segment of the image with an area of interest;

dividing the at least one segment into portions containing single characters and/or combinations of characters;

performing a validation of a plurality of single characters and/or a plurality of combinations of characters, wherein the validation is carried out for each of the single character and/or character combinations for at least two different categories;

creating score values for each category for each validated single character and/or each character combination;

creating feature vectors for each single character and/or each character combination, wherein components of the feature vectors are the score values for the single character and/or the character combination for each respective category; and

classifying the feature vectors to determine whether the single character or the character combination to which the feature vector is associated is genuine,

wherein the validation of the single characters and/or the character combinations and/or associated scoring of values for each category and/or the classification of the feature vectors is performed using an artificial neural network,

wherein the classification involves a cluster analysis,

wherein a single-character cluster analysis is performed for each feature vector associated with a corresponding single character, a multi-character cluster analysis is performed for feature vectors associated with a plurality of characters, and a document-wide cluster analysis is performed for all feature vectors associated with the characters of the document,

wherein the document-wide cluster analysis comprises obtaining a similarity indication between a feature vector associated with a single character or a combination of characters, and an aggregated mean feature vector associated with all characters in the document, and

wherein, when the similarity indication violates a defined similarity threshold, the single character associated with a corresponding dissimilar feature vector is considered to be non-genuine.

2. The method of claim 1 , wherein the at least two different categories used for the validation of the single characters and/or the character combinations include a font category, an overlay category, a background and foreground category, a font alignment category, a readability category, a completeness category, a usage of artificial filters category, a steganographic manipulation category, or a combination thereof.

3. The method of claim 2 , wherein the validation of the single characters and/or the character combinations according to the background and foreground category comprises a bonding analysis of a character in a portion.

4. The method of claim 2 , wherein the validation of the single characters and/or the character combinations according to the font alignment category comprises obtaining a distance between two adjacent characters and/or two adjacent combinations of characters.

5. The method of claim 2 , wherein the validation of the single characters and/or the character combinations according to the artificial filter category comprises passing each character through an analysis dedicated to identification of manipulation caused by artificial filter use.

6. The method of claim 2 , wherein the validation of the single characters and/or the character combinations according to the steganographic manipulation category comprises an error level analysis applied to the document, wherein the error level analysis comprises a comparison of the image with a compressed version of the image.

7. The method of claim 1 , wherein the single-character cluster analysis comprises obtaining a similarity indication between at least two feature vectors associated with single characters, wherein, when the similarity indication violates the defined similarity threshold, the single character associated with the corresponding dissimilar feature vector is considered to be non-genuine; and

wherein the multi-character cluster analysis comprises obtaining a similarity indication between at least two feature vectors associated with a combination of characters, wherein, when the similarity indication violates the defined similarity threshold, the plurality of characters associated with the corresponding dissimilar feature vector are considered to be non-genuine.

8. The method of claim 1 , wherein obtaining the similarity indication comprises calculating a cosine similarity, wherein calculating the cosine similarity comprises calculating a dot product between at least two feature vectors and a magnitude of the at least two feature vectors.

9. The method of claim 8 , wherein the defined similarity threshold lies between 0 and 1, and the defined similarity threshold is violated if the similarity indication is equal to or lower than the defined similarity threshold.

10. A computer device comprising:

at least one processor; and

at least one non-volatile memory comprising, executable instructions that, when executed by the at least one processor, cause the at least one processor to:

generate an image of a document to be audited;

pre-process the image to obtain at least one segment of the image with an area of interest;

divide the at least one segment into portions containing single characters and/or combinations of characters;

perform a validation of at least two single characters and/or at least two combinations of characters, wherein the validation is carried out for each of the single character and/or the character combinations for at least two different categories;

create score values for each category for each validated single character and/or each validated character combination;

create feature vectors for each single character and/or character combination, wherein components of the feature vectors are the score values for the single character and/or the character combination for each respective category; and

classify the feature vectors to determine whether the single character or the character combination to which the feature vector is associated is genuine,

wherein the validation of the single characters and/or the character combinations and/or associated scoring of values for each category and/or the classification of the feature vectors is performed using an artificial neural network,

wherein the classification involves a cluster analysis,

wherein a single-character cluster analysis is performed for each feature vector associated with a corresponding single character, a multi-character cluster analysis is performed for feature vectors associated with a plurality of characters, and a document-wide cluster analysis is performed for all feature vectors associated with the characters of the document,

wherein the document-wide cluster analysis comprises obtaining a similarity indication between a feature vector associated with a single character or a combination of characters, and an aggregated mean feature vector associated with all characters in the document, and

wherein, when the similarity indication violates a defined similarity threshold, the single character associated with a corresponding dissimilar feature vector is considered to be non-genuine.

11. The computer device of claim 10 , wherein the at least two different categories used for the validation of the single characters and/or the character combinations include a font category, an overlay category, a background and foreground category, a font alignment category, a readability category, a completeness category, a usage of artificial filters category, a steganographic manipulation category, or a combination thereof.

12. The computer device of claim 11 , wherein the validation of the single characters and/or the character combinations according to the background and foreground category comprises a bonding analysis of a character in a portion.

13. The computer device of claim 11 , wherein the validation of the single characters and/or the character combinations according to the font alignment category comprises obtaining a distance between two adjacent characters and/or two adjacent combinations of characters.

14. The computer device of claim 11 , wherein the validation of the single characters and/or the character combinations according to the artificial filter category comprises passing each character through an analysis dedicated to identification of manipulation caused by artificial filter use.

15. The computer device of claim 11 , wherein the validation of the single characters and/or the character combinations according to the steganographic manipulation category comprises an error level analysis applied to the document, wherein the error level analysis comprises a comparison of the image with a compressed version of the image.

16. The computer device of claim 10 , wherein the single-character cluster analysis comprises obtaining a similarity indication between at least two feature vectors associated with single characters, wherein, when the similarity indication violates the defined similarity threshold, the single character associated with the corresponding dissimilar feature vector is considered to be non-genuine; and

wherein the multi-character cluster analysis comprises obtaining a similarity indication between at least two feature vectors associated with a combination of characters, wherein, when the similarity indication violates the defined similarity threshold, the plurality of characters associated with the corresponding dissimilar feature vector are considered to be non-genuine.

17. The computer device of claim 10 , wherein obtaining the similarity indication comprises calculating a cosine similarity, wherein calculating the cosine similarity comprises calculating a dot product between at least two feature vectors and a magnitude of the at least two feature vectors.

18. The computer device of claim 17 , wherein the defined similarity threshold lies between 0 and 1, and the defined similarity threshold is violated if the similarity indication is equal to or lower than the defined similarity threshold.

19. A non-transitory computer-readable storage medium encoded with a computer program, the computer program comprising a plurality of program code instructions that when executed by one or more processors cause the one or more processors to perform operations comprising:

generate an image of a document to be audited;

pre-process the image to obtain at least one segment of the image with an area of interest;

divide the at least one segment into portions containing single characters and/or combinations of characters;

perform a validation of at least two single characters and/or at least two combinations of characters, wherein the validation is carried out for each of the single character and/or the character combinations for at least two different categories;

create score values for each category for each validated single character and/or each validated character combination;

create feature vectors for each single character and/or each character combination, wherein components of the feature vectors are the score values for the single character and/or character combination for each respective category; and

classify the feature vectors to determine whether the single character or the character combination to which the feature vector is associated is genuine,

wherein the validation of the single characters and/or the character combinations and/or associated scoring of values for each category and/or the classification of the feature vectors is performed using an artificial neural network,

wherein the classification involves a cluster analysis,

wherein a single-character cluster analysis is performed for each feature vector associated with a corresponding single character, a multi-character cluster analysis is performed for feature vectors associated with a plurality of characters, and a document-wide cluster analysis is performed for all feature vectors associated with the characters of the document,

wherein the document-wide cluster analysis comprises obtaining a similarity indication between a feature vector associated with a single character or a combination of characters, and an aggregated mean feature vector associated with all characters in the document, and

wherein, when the similarity indication violates a defined similarity threshold, the single character associated with a corresponding dissimilar feature vector is considered to be non-genuine.