DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1-7 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Kramme et al. (US 10,949,852 B1), hereafter Kramme.
Regarding claim 1, Kramme teaches a system, comprising: a computing device comprising a processor and a memory; and machine-readable instructions stored in the memory that, when executed by the processor, cause the computing device ([Col. 2, lines 5-11] “In one embodiment, a computer-implemented method of facilitating detection of document-related fraud is implemented in one or more servers. The method may include: (1) generating or updating, by one or more processors of the one or more servers, fraudulent document detection rules at least by training a machine learning program…”) to at least:
identify with a machine learning algorithm one or more unverified fields from within an unverified document associated with an entity (Kramme teaches establishing machine learning rules for a specific document type (such as a specific ID document from a specific bank) by analyzing past verified and fraudulent documents, and these rules are used to identify fraudulent and unverified fields in a new document of the same type. Figs. 3F and 5 show an overview of the method, and Fig. 4F shows examples of some rules applied to document fields including checking the font, alignment, colors, etc.);
identify a verified document corresponding to the unverified document based at least in part on the entity (See Fig. 3F. The multi-document image data 202 contains verified documents used to establish the rules for a document type. These rules are applied to new unverified documents to determine if the font, alignment, etc. matches the verified document rules. The document type can be general or specific to a particular business’s document format. [Col. 34, lines 47-54] “The physical documents and the fraudulent document detection rules may correspond to a first type of document, which may be a fairly general type (e.g., a personal check or other general type of financial instrument, or a driver's license or other general type of identification document, etc.), or a more specific type (e.g., a personal check associated with a particular bank, or a driver's license from a particular state, etc.).”);
compare the one or more unverified fields of the unverified document to one or more verified fields of the verified document (Step 206 of Fig. 3F shows comparing the unverified document 210 to the verified documents 202 using the learned rules for that document type. Referring to Fig. 5, [Col. 34, lines 55-65] “The rules generated or updated at block 302 may specify, for example, an acceptable tolerance for an absolute or relative measurement pertaining to a document (e.g., ratio of overall length to width, of various decorative or functional features depicted on the document, etc.). As other examples, the rules may specify an acceptable font or set of fonts, an acceptable pattern or set of patterns, an acceptable color or set (e.g., range) of colors, and so on. As yet another example, the rules may specify one or more acceptable field locations (e.g., placement of a signature line within the document, etc.).”); and
determine whether the unverified document is fraudulent based at least in part on a comparison of the one or more unverified fields to the one or more verified fields (Fig 4F shows counterfeit and forgery factors that are analyzed to determine if the document is fraudulent. At step 206 in Fig. 3F, the document may be flagged as fraudulent if the fields are outside of the tolerances for the applied rules shown in Fig. 4F. If flagged, then step 212 involves a manual review by a human to check for fraud. Step 214 is the final determination of a fraudulence. [Col. 25, lines 4-11] “In some embodiments, further analysis (e.g., a manual review, or further automated review using additional data sources, etc.) may be performed at a process stage 212. The additional analysis may then be used to make a final fraud determination (e.g., a final decision on whether the document is fraudulent) at process stage 214. For example, the process stage 206 may act as a filter, and flag only those documents having a relatively high probability of being fraudulent.”).
Regarding claim 2, Kramme teaches the system of claim 1, wherein the machine-readable instructions, when executed, further cause the computing device to at least: identify one or more objects from within the unverified document; and perform one or more object checks, each of the one or more object checks corresponding to a respective one of the one or more objects ([Col. 33, lines 3-11] “The factors considered under the rule set 290 may include a number of counterfeit factors 292 and a number of forgery factors 294, each of which may be evaluated by image analysis unit 52 of FIG. 1 using one or more image processing techniques. The counterfeit factors 292 may relate to the look, presentation, format and/or structure of the document, while the forgery factors 294 may relate to the substance, style or format of information entered in one or more fields of the document.”).
Regarding claim 3, Kramme teaches the system of claim 2, wherein the one or more object checks comprise at least one of an annotation check, a native document check, a duplicate object check, a modified text check, a hidden version check, a cross-reference table check, a date check, or a negative library check (Kramme teaches performing a negative library check. The multi-document image data 202 that is used for creating the fraud detection rules contains both verified and fraudulent past documents, so the rules consider characteristics of both verified fields and fraudulent fields. [Col. 23, line 64 – Col. 24, line 13] “In the process flow 200, multi-document image data 202 may represent digital images of a number (e.g., thousands) of physical documents of one or more types. The multi-document image data 202 may include data obtained (e.g., by external data collection unit 42 of FIG. 1) from merchant computing systems 22 (e.g., point-of-sale devices with cameras for document identification) and/or from FAMS 14 (e.g., images of personal checks), for example. In some embodiments, the multi-document image data 202 may only include data representing images that customers (or other individuals associated with the documents) have expressly consented to share (e.g., as a prerequisite to making a purchase, or in exchange for fraud protection services, etc.).”).
Regarding claim 4, Kramme teaches the system of claim 1, wherein the machine-readable instructions, when executed, further cause the computing device to at least: determine a respective verified font for at least one of the one or more verified fields (In Fig. 3F, the system establishes rules based on the past verified documents. These rules include setting a tolerance for an acceptable font for a new unverified document to be analyzed; see Fig. 4F.);
determine a respective unverified font for a corresponding one of the one or more unverified fields; and compare the respective unverified font to the respective verified font (If the new unverified document contains a font which is outside of the font tolerance, then the unverified document may be fraudulent. [Col. 33, lines 12-34] “As seen in FIG. 4F, the counterfeit factors 292 may include: … (4) whether one or more fonts on the document (excluding user-entered field contents) are outside one or more predetermined tolerances. …As yet another example, image analysis unit 52 may determine whether a font used for a field descriptor or other text on the document matches an expected font (e.g., by meeting a similarity threshold measured in any suitable manner).”).
Regarding claim 5, Kramme teaches the system of claim 1, wherein the machine-readable instructions, when executed, further cause the computing device to at least: determine a respective verified alignment for each of the one or more verified fields (In Fig. 3F, the system establishes rules based on the past verified documents. These rules include setting a tolerance for acceptable dimensions and angles for fields in a new unverified document to be analyzed; see Fig. 4F.);
determine a respective unverified alignment for each of the one or more unverified fields; and compare each respective unverified alignment to each respective verified alignment (If the new unverified document contains field dimensions, angles, and/or alignments which are outside of the tolerance, then the unverified document may be fraudulent. [Col. 33, lines 12-17] “As seen in FIG. 4F, the counterfeit factors 292 may include: (1) whether one or more absolute or relative dimensions and/or angles of the document, or of lines, illustrations, patterns, etc. shown on the document (excluding user-entered contents in fields such as the signature line), are outside one or more predetermined tolerances;”).
Regarding claim 6, Kramme teaches the system of claim 1, wherein the machine-readable instructions, when executed, further cause the computing device to at least: identify one or more related fields of the one or more unverified fields within the document, the one or more related fields comprising related information (In Fig. 3F, the system establishes rules based on the past verified documents. These rules include setting a tolerance for the contents within the fields; see Fig. 4F.);
determine contents corresponding to each of the one or more related fields; compare the contents for each of the one or more related fields; and verify a consistency for each of the one or more related fields based at least in part on the comparison of the contents (Many different characteristics and consistencies are checked and tolerance values are determined. For example, the system checks if fields containing a date or signature are consistent with past verified documents using the rules for that document type. Additionally, colors, alignment, fonts, and other aspects of fields are checked. [Col. 33, lines 37-48] “The forgery factors 294 may include: (1) whether a signature entered in a signature field of the document match is outside a predetermined tolerance (e.g., using any suitable signature recognition technique); (2) whether handwriting entered in one or more fields of the document is outside a predetermined tolerance (e.g., by applying a suitable handwriting recognition technique); and/or (3) whether the format of information entered by a user in one or more fields does not match an expected format (e.g., using “9.12.16” rather than the expected “9/12/2016,” as established based upon other documents known to have been populated and/or submitted by the purported applicant).”).
Regarding claim 7, Kramme teaches the system of claim 1, wherein the machine-readable instructions which, when executed, cause the computing device to determine whether the unverified document is fraudulent, further cause the computing device to at least: identify a number of fraud indicators based at least in part on the comparison of the one or more unverified fields to the one or more verified fields (Fig. 4F shows some examples of factors for analyzing unverified fields; if the fields are outside of a set tolerance, then the possibility of fraud is indicated.); and
flag the unverified document as fraudulent based at least in part on the number of fraud indicators exceeding a threshold ([Col. 33, line 56 – Col. 34, line 9] “As is also seen in FIG. 4F, each of the counterfeit factors 292 and forgery factors 294 may be associated with a particular score or weighting value... The rule set 290 may then output the total score, a normalized total score, an indication of whether the total score exceeded a threshold, a probability calculated based upon the total score, and/or some other indicator or measure of the likelihood that the document is fraudulent. Alternatively, the rule set 290 may output a separate total score, normalized score, probability, or other metric, for each of counterfeit factors 292 and forgery factors 294, with the counterfeit metric indicating the likelihood that the document is a counterfeit and the forgery metric indicating the likelihood that the document was fraudulently populated by someone other than the purported person.” [Col. 34, lines 33-37] “Referring now to FIG. 5, an exemplary computer-implemented method 300 may be used to facilitate the detection of document-related fraud, e.g., by winnowing the field of documents needing a more thorough (e.g., manual) review or investigation.”).
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 8-20 are rejected under 35 U.S.C. 103 as being unpatentable over Kramme (US 10,949,852 B1) in view of Mobley et al. (US 11,651,093 B1), hereafter Mobley.
Regarding claim 8, Kramme teaches a method, comprising: performing, by a machine learning algorithm on a computing device ([Col. 2, lines 5-11] “In one embodiment, a computer-implemented method of facilitating detection of document-related fraud is implemented in one or more servers. The method may include: (1) generating or updating, by one or more processors of the one or more servers, fraudulent document detection rules at least by training a machine learning program…”),
identifying, by the machine learning algorithm, a verified document associated with the entity, the verified document corresponding to the unverified document (See Fig. 3F. The multi-document image data 202 contains verified documents used to establish the rules for a document type. These rules are applied to new unverified documents to determine if the font, alignment, etc. matches the verified document rules. The document type can be general or specific to a particular business’s document format. [Col. 34, lines 47-54] “The physical documents and the fraudulent document detection rules may correspond to a first type of document, which may be a fairly general type (e.g., a personal check or other general type of financial instrument, or a driver's license or other general type of identification document, etc.), or a more specific type (e.g., a personal check associated with a particular bank, or a driver's license from a particular state, etc.).”);
identifying, by the machine learning algorithm, a second number of fraud indicators from within the unverified document based at least in part on a comparison of the unverified document to the verified document (Fig 4F shows counterfeit and forgery factors that are analyzed to determine if the document is fraudulent. At step 206 in Fig. 3F, the document may be flagged as fraudulent if the fields are outside of the tolerances for the applied rules shown in Fig. 4F. If flagged, then step 212 involves a manual review by a human to check for fraud. Step 214 is the final determination of a fraudulence. [Col. 25, lines 4-11] “In some embodiments, further analysis (e.g., a manual review, or further automated review using additional data sources, etc.) may be performed at a process stage 212. The additional analysis may then be used to make a final fraud determination (e.g., a final decision on whether the document is fraudulent) at process stage 214. For example, the process stage 206 may act as a filter, and flag only those documents having a relatively high probability of being fraudulent.”).
As shown above, Kramme teaches a general architecture including developing a set of rules based on past verified documents and applying the rules to flag fraudulent documents. However, the rules provided by Kramme do not include metadata analysis rules. Thus, Kramme fails to teach a metadata analysis of an unverified document associated with an entity, the metadata analysis identifying a first number of fraud indicators; and flagging, by the machine learning algorithm, the unverified document as fraudulent based at least in part on the first number of fraud indicators and the second number of fraud indicators exceeding a threshold.
However, Mobley teaches a metadata analysis of an unverified document associated with an entity, the metadata analysis identifying a first number of fraud indicators (Mobley teaches utilizing metadata-based detectors and content-based detectors for analyzing the metadata and objects of a new unverified document in comparison to past verified documents. See [Col. 5, line 4 – Col. 7, line 4] teaching examples of metadata-based detectors, and see [Col. 7, line 5 – Col. 10, line 56] teaching examples of content-based detectors.);
flagging, by the machine learning algorithm, the unverified document as fraudulent based at least in part on the first number of fraud indicators and the second number of fraud indicators exceeding a threshold (Mobley teaches combining the results of both content-based detectors (similar to Kramme’s rules) and metadata detectors when determining fraud. [Col 4. Line 62 – Col. 5, line 2] “For each document, the document authentication system 112 provides an anomaly score (0-100) predicting the probability of the document has been tampered with or forged. The anomaly score is modeled and aggregated from the feature-specific anomaly scores produced by multiple individual detectors. In one embodiment, those individual detectors include metadata-based detectors and content-based detectors.”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Kramme’s invention to include additional rules, such as metadata analysis rules, when determining if a document is fraudulent. This modification would further improve Kramme’s invention by including more rules for checking features which are not visible through image analysis alone. See the rationale to combine Kramme and Mobley in the rejection to claim 3 above for further discussion.
Regarding claim 9, Kramme and Mobley teach the method of claim 8. Kramme further teaches wherein identifying the second number of fraud indicators, further comprises: identifying, by the machine learning algorithm, one or more unverified fields from within the unverified document (Kramme teaches establishing machine learning rules for a specific document type (such as a specific ID document from a specific bank) by analyzing past verified and fraudulent documents, and these rules are used to identify fraudulent and unverified fields in a new document of the same type. Figs. 3F and 5 show an overview of the method, and Fig. 4F shows examples of some rules applied to document fields including checking the font, alignment, colors, etc.);
comparing, by the machine learning algorithm, the one or more unverified fields of the unverified document to a corresponding one or more verified fields of the verified document (Step 206 of Fig. 3F shows comparing the unverified document 210 to the verified documents 202 using the learned rules for that document type. Referring to Fig. 5, [Col. 34, lines 55-65] “The rules generated or updated at block 302 may specify, for example, an acceptable tolerance for an absolute or relative measurement pertaining to a document (e.g., ratio of overall length to width, of various decorative or functional features depicted on the document, etc.). As other examples, the rules may specify an acceptable font or set of fonts, an acceptable pattern or set of patterns, an acceptable color or set (e.g., range) of colors, and so on. As yet another example, the rules may specify one or more acceptable field locations (e.g., placement of a signature line within the document, etc.).”); and
identifying, by the machine learning algorithm, a second number of fraud indicators based at least in part on the comparison of the one or more unverified fields (Fig 4F shows counterfeit and forgery factors that are analyzed to determine if the document is fraudulent. At step 206 in Fig. 3F, the document may be flagged as fraudulent if the fields are outside of the tolerances for the applied rules shown in Fig. 4F.).
Regarding claim 10, Kramme and Mobley teach the method of claim 9. Kramme further teaches wherein comparing the one or more fields to the one or more verified fields further comprises: determining, by the machine learning algorithm, a respective verified font for at least one of the one or more verified fields (In Fig. 3F, the system establishes rules based on the past verified documents. These rules include setting a tolerance for an acceptable font for a new unverified document to be analyzed; see Fig. 4F.);
determining, by the machine learning algorithm, a respective unverified font for a corresponding one of the one or more unverified fields; and comparing, by the machine learning algorithm, the respective unverified font to the respective verified font (If the new unverified document contains a font which is outside of the font tolerance, then the unverified document may be flagged as fraudulent. [Col. 33, lines 12-34] “As seen in FIG. 4F, the counterfeit factors 292 may include: … (4) whether one or more fonts on the document (excluding user-entered field contents) are outside one or more predetermined tolerances. …As yet another example, image analysis unit 52 may determine whether a font used for a field descriptor or other text on the document matches an expected font (e.g., by meeting a similarity threshold measured in any suitable manner).”).
Regarding claim 11, Kramme and Mobley teach the method of claim 9. Kramme further teaches wherein comparing the one or more unverified fields to the one or more verified fields further comprises: determining, by the machine learning algorithm, a respective verified alignment for at least one of the one or more verified fields (In Fig. 3F, the system establishes rules based on the past verified documents. These rules include setting a tolerance for acceptable dimensions and angles for fields in a new unverified document to be analyzed; see Fig. 4F.);
determining, by the machine learning algorithm, a respective unverified alignment for a corresponding one of the one or more unverified fields; and comparing, by the machine learning algorithm, the respective unverified alignment to the respective verified alignment (If the new unverified document contains field dimensions, angles, and/or alignments which are outside of the tolerance, then the unverified document may be fraudulent. [Col. 33, lines 12-17] “As seen in FIG. 4F, the counterfeit factors 292 may include: (1) whether one or more absolute or relative dimensions and/or angles of the document, or of lines, illustrations, patterns, etc. shown on the document (excluding user-entered contents in fields such as the signature line), are outside one or more predetermined tolerances;”).
Regarding claim 12, Kramme and Mobley teach the method of claim 9. Kramme further teaches further comprising: identifying, by the machine learning algorithm, one or more related fields of the one or more unverified fields within the unverified document, the one or more related fields comprising related information (In Fig. 3F, the system establishes rules based on the past verified documents. These rules include setting a tolerance for the contents within the fields; see Fig. 4F.);
determining, by the machine learning algorithm, contents corresponding to each of the one or more related fields; comparing, by the machine learning algorithm, the contents for each of the one or more related fields; and verifying, by the machine learning algorithm, a consistency for each of the one or more related fields based at least in part on the comparison of the contents (Many different characteristics and consistencies are checked and tolerance values are determined. For example, the system checks if fields containing a date or signature are consistent with past verified documents using the rules for that document type. Additionally, colors, alignment, fonts, and other aspects of fields are checked. [Col. 33, lines 37-48] “The forgery factors 294 may include: (1) whether a signature entered in a signature field of the document match is outside a predetermined tolerance (e.g., using any suitable signature recognition technique); (2) whether handwriting entered in one or more fields of the document is outside a predetermined tolerance (e.g., by applying a suitable handwriting recognition technique); and/or (3) whether the format of information entered by a user in one or more fields does not match an expected format (e.g., using “9.12.16” rather than the expected “9/12/2016,” as established based upon other documents known to have been populated and/or submitted by the purported applicant).”).
Regarding claim 13, Kramme and Mobley teach the method of claim 8. Mobley further teaches wherein performing the metadata analysis of the unverified document further comprises: identifying, by the machine learning algorithm, one or more objects from the unverified document; and performing, by the machine learning algorithm, one or more checks, each of the one or more checks corresponding to a respective one of the one or more objects (Mobley teaches utilizing metadata-based detectors and content-based detectors for analyzing the metadata and objects of a new unverified document in comparison to past verified documents. See [Col. 5, line 4 – Col. 7, line 4] teaching examples of metadata-based detectors, and see [Col. 7, line 5 – Col. 10, line 56] teaching examples of content-based detectors.).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Kramme’s invention to include additional rules, such as metadata analysis rules, when determining if a document is fraudulent. This modification would further improve Kramme’s invention by including more rules for checking features which are not visible through image analysis alone. See the rationale to combine Kramme and Mobley in the rejection to claim 3 above for further discussion.
Regarding claim 14, Kramme and Mobley teach the method of claim 8. Kramme further teaches further comprising: identifying, by the machine learning algorithm, one or more patterns from within the unverified document; calculating, by the machine learning algorithm, a pattern score based at least in part on the one or more patterns identified; identifying, by the machine learning algorithm, a third number of fraud indicators based at least in part on the pattern score exceeding a threshold (Kramme teaches several rules for analyzing patterns of lines, colors, alignment, etc. of a new unverified document. [Col. 33, lines 12-30] “As seen in FIG. 4F, the counterfeit factors 292 may include: (1) whether one or more absolute or relative dimensions and/or angles of the document, or of lines, illustrations, patterns, etc. shown on the document (excluding user-entered contents in fields such as the signature line), are outside one or more predetermined tolerances; (2) whether one or more colors on the document are outside a predetermined tolerance (e.g., color/frequency range); (3) whether one or more line thicknesses of the document (excluding user-entered field contents) are outside one or more predetermined tolerances… For example, image analysis unit 52 may determine whether the ratio of the document length to the document width is within 0.1% of an expected value. As another example, image analysis unit 52 may determine whether horizontal and vertical lines on the document are within 0.3 degrees of the horizontal and vertical edges of the document, respectively.”).
Although Kramme teaches the second and third number of fraud indicators, Kramme does not teach performing a metadata analysis for generating the first number of fraud indicators. Thus, Kramme fails to teach flagging, by the machine learning algorithm, the unverified document as fraudulent based at least in part on the first number of fraud indicators, the second number of fraud indicators, and the third number of fraud indicators exceeding a threshold.
However, Mobley teaches flagging, by the machine learning algorithm, the unverified document as fraudulent based at least in part on the first number of fraud indicators, the second number of fraud indicators, and the third number of fraud indicators exceeding a threshold (Fig. 1 shows that the results of all detectors are aggregated by the aggregator 106 to determine a final document anomaly score 110. The detectors include both metadata analysis and content analysis. [Col. 3, lines 37-40] “In one implementation, target documents that produce target document anomaly scores that exceed a certain threshold are flagged as ‘suspect’.” Furthermore, Kramme also teaches comparing the fraud indicators to thresholds when making a determination. See Kramme [Col. 24, line 58 – Col. 25, line 22]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Kramme’s invention to include additional rules, such as metadata analysis rules, when determining if a document is fraudulent. This modification would further improve Kramme’s invention by including more rules for checking features which are not visible through image analysis alone. See the rationale to combine Kramme and Mobley in the rejection to claim 3 above for further discussion.
Regarding claim 15, Kramme teaches a non-transitory, computer-readable medium, comprising machine-readable instructions that, when executed by a processor of a computing device ([Col. 2, lines 53-57] “In another embodiment, a non-transitory, computer-readable medium stores instructions that, when executed by one or more processors, cause the one or more processors to: (1) generate or update fraudulent document detection rules at least by training a machine learning program…”), cause the computing device to at least:
identify a verified document associated with the entity, the verified document corresponding to the unverified document (See Fig. 3F. The multi-document image data 202 contains verified documents used to establish the rules for a document type. These rules are applied to new unverified documents to determine if the font, alignment, etc. matches the verified document rules. The document type can be general or specific to a particular business’s document format. [Col. 34, lines 47-54] “The physical documents and the fraudulent document detection rules may correspond to a first type of document, which may be a fairly general type (e.g., a personal check or other general type of financial instrument, or a driver's license or other general type of identification document, etc.), or a more specific type (e.g., a personal check associated with a particular bank, or a driver's license from a particular state, etc.).”);
identify a second number of fraud indicators from within the unverified document based at least in part on a comparison of the unverified document to the verified document (Fig 4F shows counterfeit and forgery factors that are analyzed to determine if the document is fraudulent. At step 206 in Fig. 3F, the document may be flagged as fraudulent if the fields are outside of the tolerances for the applied rules shown in Fig. 4F. If flagged, then step 212 involves a manual review by a human to check for fraud. Step 214 is the final determination of a fraudulence. [Col. 25, lines 4-11] “In some embodiments, further analysis (e.g., a manual review, or further automated review using additional data sources, etc.) may be performed at a process stage 212. The additional analysis may then be used to make a final fraud determination (e.g., a final decision on whether the document is fraudulent) at process stage 214. For example, the process stage 206 may act as a filter, and flag only those documents having a relatively high probability of being fraudulent.”).
As shown above, Kramme teaches a general architecture including developing a set of rules based on past verified documents and applying the rules to flag fraudulent documents. However, the rules provided by Kramme do not include metadata analysis rules. Thus, Kramme fails to teach performing a metadata analysis of an unverified document associated with an entity, the metadata analysis identifying a first number of fraud indicators; and flagging, by the machine learning algorithm, the unverified document as fraudulent based at least in part on the first number of fraud indicators and the second number of fraud indicators exceeding a threshold.
However, Mobley teaches causing the computing device to perform a metadata analysis of an unverified document associated with an entity, the metadata analysis identifying a first number of fraud indicators (Mobley teaches utilizing metadata-based detectors and content-based detectors for analyzing the metadata and objects of a new unverified document in comparison to past verified documents. See [Col. 5, line 4 – Col. 7, line 4] teaching examples of metadata-based detectors, and see [Col. 7, line 5 – Col. 10, line 56] teaching examples of content-based detectors.);
flag the unverified document as fraudulent based at least in part on the first number of fraud indicators and the second number of fraud indicators exceeding a threshold (Mobley teaches combining the results of both content-based detectors (similar to Kramme’s rules) and metadata detectors when determining fraud. [Col 4. Line 62 – Col. 5, line 2] “For each document, the document authentication system 112 provides an anomaly score (0-100) predicting the probability of the document has been tampered with or forged. The anomaly score is modeled and aggregated from the feature-specific anomaly scores produced by multiple individual detectors. In one embodiment, those individual detectors include metadata-based detectors and content-based detectors.”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Kramme’s invention to include additional rules, such as metadata analysis rules, when determining if a document is fraudulent. This modification would further improve Kramme’s invention by including more rules for checking features which are not visible through image analysis alone. See the rationale to combine Kramme and Mobley in the rejection to claim 3 above for further discussion.
Regarding claim 16, Kramme and Mobley teach the non-transitory, computer-readable medium of claim 15. Kramme further teaches wherein the machine-readable instructions, when executed by the processor, further cause the computing device to at least: identify one or more objects from the unverified document; and perform one or more checks, each of the one or more checks corresponding to a respective one of the one or more objects ([Col. 33, lines 3-11] “The factors considered under the rule set 290 may include a number of counterfeit factors 292 and a number of forgery factors 294, each of which may be evaluated by image analysis unit 52 of FIG. 1 using one or more image processing techniques. The counterfeit factors 292 may relate to the look, presentation, format and/or structure of the document, while the forgery factors 294 may relate to the substance, style or format of information entered in one or more fields of the document.”).
Regarding claim 17, Kramme and Mobley teach the non-transitory, computer-readable medium of claim 15. Kramme further teaches wherein the machine-readable instructions which, when executed by the processor, cause the computing device to identify a second number of fraud indicators from within the unverified document, further cause the computing device to at least: identify one or more unverified fields from within the unverified document (Kramme teaches establishing machine learning rules for a specific document type (such as a specific ID document from a specific bank) by analyzing past verified and fraudulent documents, and these rules are used to identify fraudulent and unverified fields in a new document of the same type. Figs. 3F and 5 show an overview of the method, and Fig. 4F shows examples of some rules applied to document fields including checking the font, alignment, colors, etc.);
compare the one or more unverified fields of the unverified document to a corresponding one or more verified fields of the verified document (Step 206 of Fig. 3F shows comparing the unverified document 210 to the verified documents 202 using the learned rules for that document type. Referring to Fig. 5, [Col. 34, lines 55-65] “The rules generated or updated at block 302 may specify, for example, an acceptable tolerance for an absolute or relative measurement pertaining to a document (e.g., ratio of overall length to width, of various decorative or functional features depicted on the document, etc.). As other examples, the rules may specify an acceptable font or set of fonts, an acceptable pattern or set of patterns, an acceptable color or set (e.g., range) of colors, and so on. As yet another example, the rules may specify one or more acceptable field locations (e.g., placement of a signature line within the document, etc.).”); and
identify a second number of fraud indicators based at least in part on the comparison of the one or more unverified fields (Fig 4F shows counterfeit and forgery factors that are analyzed to determine if the document is fraudulent. At step 206 in Fig. 3F, the document may be flagged as fraudulent if the fields are outside of the tolerances for the applied rules shown in Fig. 4F.).
Regarding claim 18, Kramme and Mobley teach the non-transitory, computer-readable medium of claim 15. Kramme further teaches wherein the machine-readable instructions, when executed by the processor, further cause the computing device to at least: identify one or more patterns from within the unverified document; calculate a pattern score based at least in part on the one or more patterns identified; and identify a third number of fraud indicators based at least in part on the pattern score exceeding a threshold (Kramme teaches several rules for analyzing patterns of lines, colors, alignment, etc. of a new unverified document. [Col. 33, lines 12-30] “As seen in FIG. 4F, the counterfeit factors 292 may include: (1) whether one or more absolute or relative dimensions and/or angles of the document, or of lines, illustrations, patterns, etc. shown on the document (excluding user-entered contents in fields such as the signature line), are outside one or more predetermined tolerances; (2) whether one or more colors on the document are outside a predetermined tolerance (e.g., color/frequency range); (3) whether one or more line thicknesses of the document (excluding user-entered field contents) are outside one or more predetermined tolerances… For example, image analysis unit 52 may determine whether the ratio of the document length to the document width is within 0.1% of an expected value. As another example, image analysis unit 52 may determine whether horizontal and vertical lines on the document are within 0.3 degrees of the horizontal and vertical edges of the document, respectively.”).
Regarding claim 19, Kramme and Mobley teach the non-transitory, computer-readable medium of claim 15. Mobley further teaches wherein the first number of fraud indicators comprises at least one of an annotation tag, a hidden version, or a modified text field ([Col. 9, lines 15-36] teaches detecting when annotations have been added to a PDF. [Col. 9, lines 45-50] teaches checking for previous version information in the PDF code. [Col. 8, lines 45-53 teaches checking for text that has been injected at a later time than other text within the document).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Kramme’s invention to include additional rules, such as metadata analysis rules, when determining if a document is fraudulent. This modification would further improve Kramme’s invention by including more rules for checking features which are not visible through image analysis alone. See the rationale to combine Kramme and Mobley in the rejection to claim 3 above for further discussion.
Regarding claim 20, Kramme and Mobley teach the non-transitory, computer-readable medium of claim 15. Kramme further teaches wherein the second number of fraud indicators comprises at least one of a font inconsistency, an alignment inconsistency, or a content inconsistency (In Fig. 3F, the system establishes rules based on the past verified documents. These rules include setting a tolerance for an acceptable font for a new unverified document to be analyzed and setting a tolerance for acceptable dimensions and angles for fields in a new unverified document to be analyzed. Furthermore, colors, alignment, fonts, and other aspects of fields are checked. See [Col. 33, lines 37-48].).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Balakrishnan et al. (US 2021/0124919 A1) teaches systems and methods for determining if documents are fraudulent by comparing a new unverified document to a template for that type of document. The methods involve comparing fonts, layouts, etc. between the document and the template.
Berger et al. (US 2020/0110932 A1) teaches systems and methods for detecting a fraudulent document by measuring the dissimilarity between a new unverified document and known, registered documents.
Kumar et al. (Digital Verification: An Efficient Fraud Document Detection for Insurance Claims using IoT and ML Approaches. ICPCSN. pp. 242-247.) teaches methods of using machine learning to identify fraudulent certificates.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ERIC JAMES SHOEMAKER whose telephone number is (571)272-6605. The examiner can normally be reached Monday through Friday from 8am to 5pm ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner' s supervisor, JENNIFER MEHMOOD, can be reached at (571)272-2976. The fax phone number for the organization where this application or proceeding is assigned is (571)273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Eric Shoemaker/
Patent Examiner
/JENNIFER MEHMOOD/Supervisory Patent Examiner, Art Unit 2664
Prosecution Insights