Character authenticity determination

1. A computer-implemented method for assessing if a sample image comprises one or more characters from a predefined selection of characters, the method comprising:

passing the sample image through an alignment network to form a transformed image, the alignment network being configured to align images of characters within the sample image with images of characters from the predefined selection of characters, and thereby generate the transformed image;

computing a similarity of one or more characters within the transformed image with one or more corresponding reference images of characters from the predefined selection of characters to form a similarity score of characters in the sample image with the characters of the predefined selection of characters; and

declaring the sample image not to comprise at least one character from the predefined selection of characters if the similarity score indicates a significant enough difference.

2. The computer-implemented method of claim 1 wherein the step of passing a sample image through an alignment network to form a transformed image comprises:

processing the sample image with the alignment network to form a corrective transformation; and

applying the corrective transformation to the sample image to form a transformed image.

3. The computer-implemented method of claim 1 wherein the alignment network comprises a conditional alignment manifold.

4. The computer-implemented method of claim 3 further comprising:

forming the conditional alignment manifold using an unsupervised machine learning algorithm.

5. The computer-implemented method of claim 4 wherein the unsupervised machine learning algorithm comprises a densely fused spatial transformer network.

6. The computer-implemented method of claim 4 wherein the step of forming the conditional alignment manifold further comprises:

training with training images comprising a selection of characters from a predefined training selection of characters.

7. The computer-implemented method of claim 6 wherein each training image comprises a character from the predefined training selection of characters.

8. The computer-implemented method of claim 6 wherein the predefined training selection of characters consists of all characters that are associated with a section of an official document.

9. The computer-implemented method of claim 1 wherein the predefined selection of characters consists of all characters that are associated with a given section of an official document.

10. The computer-implemented method of claim 8 wherein all characters that are associated with a section of an official document comprises

all upper case characters and a chevron character of a font used in a machine readable zone of an official document.

11. The computer-implemented method of claim 1 wherein the step of computing a similarity of the transformed image with a corresponding reference image from the predefined selection of characters to form a similarity score further comprises a first similarity testing method that comprises:

performing optical character recognition on the sample image to recognise a sample character;

selecting a matching image that comprises the sample character from the predefined selection of characters; and

computing the similarity of the transformed image with the matched image to form a similarity score.

12. The computer-implemented method of claim 1 wherein the sample image is: an image of a single character, or an image comprising multiple characters.

13. A computer-implemented method for testing the authenticity of an official document, the method comprising:

obtaining an official document image of an official document;

identifying one or more characters in the official document image;

for at least one identified character, segmenting the official document image to form at least one sample image, each sample image comprising an identified character;

for each sample image, assessing if the sample image comprises a character from a predefined selection of characters using the computer-implemented method of claim 1 ; and

declaring the official document not to be genuine if a set number of sample images are declared not to comprise a character from the predefined selection of characters,

wherein the predefined selection of characters is a selection of characters associated with at least a part of the official document.

14. The computer-implemented method of claim 13 wherein the set number of sample images is one sample image.

15. The computer-implemented method of claim 13 wherein the predefined training selection of characters is the selection of characters associated with at least a part of the official document.

16. A computer-readable medium comprising computer executable instructions for performing the method of claim 1 .

17. A computer comprising a processor configured to execute executable code stored in memory wherein the executable code comprises instructions for performing the method of claim 1 .

18. The computer-implemented method of claim 1 wherein declaring the sample image not to comprise at least one character from the predefined selection of characters if the similarity score indicates a significant enough difference comprises:

evaluating if the similarity score is less than, or greater than, than a predefined threshold score, and if so declaring the sample image not to comprise at least one character from the predefined selection of characters.

19. A computer-implemented method for forming an alignment network for a predefined training selection of characters, the method comprising:

acquiring a plurality of training images, each training image comprising a character from a predefined training selection of characters; and

for each training image, performing the following steps

passing a training image through a densely fused spatial transformer network to form a transformed training image;

computing the similarity of the transformed training image with a corresponding template image of a character from the predefined training selection of characters to form an error signal; and

training the densely fused spatial transformer network with the error signal,

wherein, after the plurality of training images has been processed, the densely fused spatial transformer network comprises an alignment network for the predefined training selection of characters.