Systems and methods for classifying payment documents during mobile image processing
Systems and methods are provided for processing an image of a financial payment document captured using a mobile device and classifying the type of payment document in order to extract the content therein. These methods may be implemented on a mobile device or a central server, and can be used to identify content on the payment document and determine whether the payment document is ready to be processed by a business or financial institution. The system can identify the type of payment document by identifying features on the payment document and performing a series of steps to determine probabilities that the payment document belongs to a specific document type. The identification steps are arranged starting with the fastest step in order to attempt to quickly determine the payment document type without requiring lengthy, extensive analysis.
1. A system comprising:
a feature knowledge database configured to store a plurality of document models generated from feature information for a plurality of document types, wherein each document model comprises probabilistic indications associated with each feature, wherein the probabilistic indications indicate the likelihood that a document is a certain type of document; and
one or processors configured to execute instructions, the one or more processors communicatively coupled with the feature knowledge database, the instructions configured to cause the processor to
receive an image of a new document,
extract feature information for a plurality of features from the image,
compare the feature information to the feature information included in the plurality of document models, wherein the features are compared according to a hierarchical classification algorithm,
for each feature being compared, assess a probability that the new document is one of the plurality of document types based on the probabilistic indications for the given feature and prior features that have been extracted and compared, and
when the probability that the new document is the one document type exceeds a threshold, then determine that the new document is that one document type.
2. The system of claim 1 , wherein the system is implemented on a mobile device.
3. The system of claim 1 , wherein the document types are related to types of checks.
4. The system of claim 3 , wherein the feature information is related to at least some of a magnetic ink character recognition (MICR) line, codeline, one or more barcodes, check number, and lock icon.
5. The system of claim 3 , wherein the feature information is related to geometric features.
6. The system of claim 5 , wherein the geometric features include at least some of width, height, and aspect ratio.
7. The system of claim 5 , wherein the instructions are further configured to cause the one or more processors to use the geometric features to eliminate certain document models of the plurality of document models.
8. The system of claim 4 , wherein the instructions are further configured to cause the one or more processors to identify a presence of the lock icon and a location of the lock icon, and to identify a corresponding document type based on the identification of the presence and location of the lock icon.
9. The system of claim 8 , wherein the instructions are further configured to cause the one or more processors to then determine a presence and type-specific properties of the MICR line, if present, and to identify the corresponding document type based on the determination of the presence and type-specific properties of the MICR line.
10. The system of claim 1 , wherein the instructions are further configured to cause the one or more processors to determine that the new document is fraudulent based on the probabilistic indications for the given feature and prior features that have been extracted and compared.
11. A method comprising using at least one hardware processor to:
maintain a feature knowledge database configured to store a plurality of document models generated from feature information for a plurality of document types, wherein each document model comprises probabilistic indications associated with each feature, wherein the probabilistic indications indicate the likelihood that a document is a certain type of document;
receive an image of a new document;
extract feature information for a plurality of features from the image;
compare the feature information to the feature information included in the plurality of document models, wherein the features are compared according to a hierarchical classification algorithm;
for each feature being compared, assess a probability that the new document is one of the plurality of document types based on the probabilistic indications for the given feature and prior features that have been extracted and compared; and
when the probability that the new document is the one document type exceeds a threshold, then determine that the new document is that one document type.
12. The method of claim 11 , wherein the document types are related to types of checks.
13. The method of claim 11 , further comprising using the at least one hardware processor to identify a presence of a lock icon and a location of the lock icon, and to identify a corresponding document type based on the identification of the presence and location of the lock icon.
14. The method of claim 11 , further comprising using the at least one hardware processor to determine a presence and type-specific properties of a magnetic ink character recognition (MICR) line, if present, and to identify a corresponding document type based on the determination of the presence and type-specific properties of the MICR line.
15. The method of claim 11 , further comprising using the at least one hardware processor to determine that the new document is fraudulent based on the probabilistic indications for the given feature and prior features that have been extracted and compared.
16. A non-transitory computer-readable medium having instructions stored thereon, wherein the instructions, when executed by a processor, cause the processor to:
maintain a feature knowledge database configured to store a plurality of document models generated from feature information for a plurality of document types, wherein each document model comprises probabilistic indications associated with each feature, wherein the probabilistic indications indicate the likelihood that a document is a certain type of document;
receive an image of a new document;
extract feature information for a plurality of features from the image;
compare the feature information to the feature information included in the plurality of document models, wherein the features are compared according to a hierarchical classification algorithm;
for each feature being compared, assess a probability that the new document is one of the plurality of document types based on the probabilistic indications for the given feature and prior features that have been extracted and compared; and
when the probability that the new document is the one document type exceeds a threshold, then determine that the new document is that one document type.
17. The non-transitory computer-readable medium of claim 16 , wherein the document types are related to types of checks.
18. The non-transitory computer-readable medium of claim 16 , wherein the instructions further cause the processor to identify a presence of a lock icon and a location of the lock icon, and to identify a corresponding document type based on the identification of the presence and location of the lock icon.
19. The non-transitory computer-readable medium of claim 16 , wherein the instructions further cause the processor to determine a presence and type-specific properties of a magnetic ink character recognition (MICR) line, if present, and to identify a corresponding document type based on the determination of the presence and type-specific properties of the MICR line.
20. The non-transitory computer-readable medium of claim 16 , wherein the instructions further cause the processor to determine that the new document is fraudulent based on the probabilistic indications for the given feature and prior features that have been extracted and compared.