Deep learning stack used in production to prevent exfiltration of image-borne identification documents

1. A method of detecting identification documents in images, referred to as image-borne identification documents, and protecting against loss by image exfiltration of the identification documents, including:

using a trained deep learning (abbreviated DL) stack to classify at least one production image by inference as containing a sensitive image-borne identification document;

wherein the trained production DL stack is configured with parameters determined using labelled ground truth data for the image-borne identification documents and examples of other image documents; and

wherein the trained DL stack is further configured to include at least a first set of layers closer to an input layer and a second set of layers further from the input layer, wherein the first set of layers were pre-trained to perform image recognition before exposing the second set of layers of the DL stack to the labelled ground truth data for the image-borne identification documents and examples of other image documents; and

using the inferred classification of the sensitive image-borne identification document in a document loss prevention system to protect against loss by image exfiltration.

2. The method of claim 1 , further including for private image-borne identification documents, capturing features produced as output from the first set of layers and retaining the captured features together with respective ground truth labels, thereby eliminating any need to retain images of the private image-borne identification documents.

3. The method of claim 1 , further including restricting training by backward propagation using the labelled ground truth data for the image-borne identification documents and the examples of other image documents to training of parameters in the second set of layers.

4. The method of claim 1 , wherein optical character recognition (abbreviated OCR) analysis of images is applied to label the images as identification documents or non-identification documents.

5. The method of claim 1 , wherein a first set of the image-borne identification documents is distorted in perspective to produce a second set of the image-borne identification documents and combining the first and second sets with the labelled ground truth data when training the DL stack by back propagation.

6. The method of claim 1 , wherein a first set of the image-borne identification documents is distorted by noise to produce a third set of the image-borne identification documents and combining the first and third sets of the image-borne identification documents with the labelled ground truth data when training the DL stack by back propagation.

7. The method of claim 1 , wherein a first set of the image-borne identification documents is distorted in focus to produce a fourth set of the image-borne identification documents and combining the first and fourth sets of the image-borne identification documents with the labelled ground truth data when training the DL stack by back propagation.

8. A tangible non-transitory computer readable storage media, including program instructions loaded into memory that, when executed on processors, cause the processors to implement a method of detecting identification documents in images, referred to as image-borne identification documents, and protecting against loss of the image-borne identification documents, the method including:

using a trained deep learning (abbreviated DL) stack to classify at least one production image by inference as containing a sensitive image-borne identification document;

wherein the trained production DL stack is configured with parameters determined using labelled ground truth data for the image-borne identification documents and examples of other image documents; and

wherein the trained DL stack is further configured to include at least a first set of layers closer to an input layer and a second set of layers further from the input layer, wherein the first set of layers were pre-trained to perform image recognition before exposing the second set of layers of the DL stack to the labelled ground truth data for the image-borne identification documents and examples of other image documents; and

using the inferred classification of the sensitive image-borne identification document in a document loss prevention system to protect against loss by image exfiltration.

9. The tangible non-transitory computer readable storage media of claim 8 , further including for private image-borne identification documents, capturing features produced as output from the first set of layers and retaining the captured features together with respective ground truth labels, thereby eliminating any need to retain images of the private image-borne identification documents.

10. The tangible non-transitory computer readable storage media of claim 8 , further including restricting training by backward propagation using the labelled ground truth data for the image-borne identification documents and the examples of other image documents to training of parameters in the second set of layers.

11. The tangible non-transitory computer readable storage media of claim 8 , wherein optical character recognition (abbreviated OCR) analysis of images is applied to label the images as identification documents or non-identification documents.

12. The tangible non-transitory computer readable storage media of claim 8 , wherein a first set of the image-borne identification documents is distorted in perspective to produce a second set of the image-borne identification documents and combining the first and second sets with the labelled ground truth data when training the DL stack by back propagation.

13. The tangible non-transitory computer readable storage media of claim 8 , wherein a first set of the image-borne identification documents is distorted by noise to produce a third set of the image-borne identification documents and combining the first and third sets of the image-borne identification documents with the labelled ground truth data when training the DL stack by back propagation.

14. A system for detecting identification documents in images, referred to as image-borne identification documents, and protecting against loss of the image-borne identification documents, the system including a processor, memory coupled to the processor, and computer instructions from the non-transitory computer readable storage media of claim 8 loaded into the memory.

15. The system of claim 14 , further including for private image-borne identification documents, capturing features produced as output from the first set of layers and retaining the captured features together with respective ground truth labels, thereby eliminating any need to retain images of the private image-borne identification documents.

16. The system of claim 14 , further including restricting training by backward propagation using the labelled ground truth data for the image-borne identification documents and the examples of other image documents to training of parameters in the second set of layers.

17. The system of claim 14 , wherein optical character recognition (abbreviated OCR) analysis of images is applied to label the images as identification documents or non-identification documents.

18. The system of claim 14 , wherein a first set of the image-borne identification documents is distorted in perspective to produce a second set of the image-borne identification documents and combining the first and second sets with the labelled ground truth data when training the DL stack by back propagation.

19. The system of claim 14 , wherein a first set of the image-borne identification documents is distorted by noise to produce a third set of the image-borne identification documents and combining the first and third sets of the image-borne identification documents with the labelled ground truth data when training the DL stack by back propagation.

20. The system of claim 14 , wherein a first set of the image-borne identification documents is distorted in focus to produce a fourth set of the image-borne identification documents and combining the first and fourth sets of the image-borne identification documents with the labelled ground truth data when training the DL stack by back propagation.