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 .
Response to Arguments
Applicants arguments/amendments filed on 1/23/2026 have been entered and made of record.
Applicant's arguments filed 1/23/2026 with respect to U.S.C. 101 have been fully considered but they are not persuasive.
Applicant argues with respect to step 2a prong one In view of the reasoning presented below, the Applicant respectfully submits that amended independent claim 1 does not fall within the category of abstract idea under the mental process grouping.
“The Applicant respectfully submits that amended independent claim 1 does not recite a mental process, a mathematical concept, or a method of organizing human activity as those groupings are defined in the 2019 Revised Patent Subject Matter Eligibility Guidance (hereinafter "Guidance"). The Office Action's 35 U.S.C. § 101 analysis treated the recited steps such as "receiving an input image," "performing content extraction," "selecting a template," "determining a confidence score," "postprocessing," "validating," and "classifying" as susceptible to performance "in the mind," thereby characterizing the claim as a "mental process." That characterization cannot be reconciled with the claim's machine-implemented and data-dependent operations that require non-trivial computation and model inference over document images and predicted templates, followed by postprocessing and a tri-part validation prior to automated classification for a straight through processing or manual annotation. The Applicant's as-filed specification repeatedly teaches that the pipeline is implemented by a unified document classifier including a model training engine, model prediction engine, and document importer that process images, execute machine learning inference, and enforce validation logic which cannot be carried out mentally or with pen and paper. The claimed disclosure confirms that the claimed method operates in a computing environment manipulating images, embeddings, confidence scores, and validation criteria at scale, and does not fall within the mental processes grouping.
Applicants argument “machine-implemented and data-dependent operations that require non-trivial computation and model inference over document images and predicted templates, followed by postprocessing and a tri-part validation prior to automated classification for a straight through processing or manual annotation.” The examiner note that machine implemented data driven” abstract ideas are still abstract. For example, simply implementing an abstract idea with a generic computer does not make an abstract idea not abstract. There is no “model training engine” claimed in claim 1 (the examiner notes that claim 2 which includes training aspects is not rejected). Generic concepts like validation processing image and inference are all can be abstract. Machine learning in the claim only limits the claim to the generic field of machine learning. No specific machine learning implementation is claimed. As such the examiner believes the above recited elements to be abstract.
Applicant further argues:
Amended independent claim 1's two classification outcomes: classifying the document for a straight through processing only upon successful validation corresponding to each of the image quality validation, the refined content extraction validation, and the layout validation, and classifying the document for a manual annotation upon unsuccessful validation, are machine-implemented decisions that rely on image quality scoring, refined content validation after template-based postprocessing, and layout integrity checks. As described in the Applicant's as-filed specification's method flow (e.g., steps 216-222 and associated text), image quality validation uses deep learning on OCR outputs to grade legibility and orientation; refined content extraction validation tests the extracted content of constant/discrete/variable fields against ranges, patterns, and consistency rules; and layout validation verifies expected field alignment, bounds, and structural conformity to the predicted template. These validations constitute technical operations that demand computational analysis of image-derived data and template geometry; they cannot be reasonably performed by observation or judgment alone without the claimed machine learning and validation logic. The conjunctive requirement: successful validation corresponding to each validation prior to a straight through processing classification, further confirms a non-mental, data-flow-driven process.
The examiner disagrees elements like classifying and validating have long been preformed as mental processes by humans. Elements like “deep learning” are not in the claim. Further the elements are described at a high level and do not require specific elements that “cannot be reasonably performed by observation or judgment alone without the claimed machine learning and validation logic”. Its not clear what specific part applicant believes cannot be performed mentally. Particularly the element “The conjunctive requirement: successful validation corresponding to each validation prior to a straight through processing classification, further confirms a non-mental, data-flow-driven process.” is abstract. This element is just a vague reference to different types of validation. A human could easily do these by looking at the image. For example quality validation could simply mean looking at the content and deciding whether in the users opinion it is quality or not. There are few no specifics on how this “validation“ is performed in the claim. A similar argument applies to the other forms of validation. It is not clear to the examiner why exactly applicant believes these elements can not be performed in the mind.
Applicant further argues:
Further, the Office Action's mental-process characterization asserted that, "under the
broadest reasonable interpretation," steps such as "receiving,""performing content extraction,""selecting,""determining,""postprocessing,""validating," and "classifying" can be performed in the mind, absent more than generic computer components. That assertion overlooks that each claimed step is defined and constrained by image-level and template-level operations, ML inference, postprocessing tied to the predicted template, and explicit, technical validations that serve as gates to a specific classification outcome. Further, as described above, image quality validation assigns a quantitative score to OCR outputs and applies orientation/legibility criteria; refined content extraction validation checks constant/discrete/variable fields against ranges, patterns, and consistency metrics after postprocessing; and layout validation enforces that field alignment and bounds match the predicted template's structure, all of which are algorithmic tasks intended for computers and applied to images and structured templates On that basis, said claim cannot be performed practically "in the mind," and therefore is not directed to a mental process
The examiner notes that humans have can perform image level and template level operations mentally. “Postprocessing” is a vague term that can also be performed mentally. Validation is a vague concept that also there is little detail in the claimed validations. Humans can easily mentally validate things. Humans can assign quantitative scores. Humans can compare to a template. The examiner notes that the examiner did not consider the concept of machine learning to be abstract but as claimed in claim 1 it does not sufficient to integrate into a practical applicant or provide significantly more. In the above arguments with respect to prong 1 applicant seems to be merely listing elements and the claiming they cannot be mentally be performed. There is not specific argument to a specific feature detailing why exactly applicant feels these elements cannot be performed in the mind. The examiner is unsure why applicant believes any of the element articulated as abstract by the examiner cannot be mentally performed. Concept like"receiving an input image," "performing content extraction," "selecting a template," "determining a confidence score," "postprocessing," "validating," and "classifying" are well within the bounds of mental processes. Unless they require specific steps which cannot be performed in the mind they are considered a mental process.
Applicant further argues
Moreover, the classification for a straight through processing is only reached upon successful validation corresponding to each validation. This conjunctive gate is not an abstract "if-then" preference but a technical requirement that guarantees data fitness and layout integrity before non-human straight through processing automation continues; if validation is unsuccessful, the method classifies the document for manual annotation and consumes correction input for learning. This dual classification outcome is a machine control decision in a computing pipeline, not a mental determination. The Applicant's as-filed specification depicts these flows in Figures 2 and 5-6 and associated text explaining pipeline components and decision gates.
Further, the Applicant submits that support for the claimed validation and classification features appears throughout the Applicant's as-filed specification: the tri-part validation logic and the straight through processing/manual annotation classification are expressly taught at steps 216-222, including the image quality validation, refined content extraction validation, and layout validation, followed by classifying for a straight through processing or classifying for manual annotation when validation fails, and then receiving correction input and learning via reinforcement learning. The Summary and system description further anchor that the method is implemented by processors and memories to realize ML inference, template selection, postprocessing, validations, and classification decisions. In summary, steps 216-222 confirm that the claim is technical and computer-implemented throughout.
In view of the above, the Applicant respectfully submits that amended independent claim 1 includes features that cannot be practically performed mentally, and is therefore not directed to an abstract idea under Prong One of Step 2A. In fact, as per the reasoning presented above, amended independent claim 1 recites features that go beyond an abstract idea.
Amended independent claim 8 recites some features that are similar to the features of amended independent claim 1. Hence, the remarks presented above for amended independent claim 1 also apply to amended independent claim 8.
Again the examiner disagrees. The examiner is not clear what applicant is trying to say with these arguments. Humans regularly perform processes like classification and validation mentally. Its not clear to the examiner why applicant believes these features cannot be performed mentally. For example, why does applicant think “dual classification” cannot be performed mentally? That’s merely classifying into two categories and can easily be performed mentally. It appears applicant its stating that these elements cannot be performed mentally because they are disclosed as being performed by a computer. Mere implementation of an abstract idea using a computer and machine learning is not sufficient to make an abstract idea not abstract.
Applicant argues with respect to 2a prong 2:
Even assuming, arguendo, that some portion of previously presented independent claim 1 could be viewed as reciting an abstract concept, amended independent claim 1 integrates any such concept into a practical application. The Guidance recognizes that features which reflect an "improvement in the functioning of a computer or an improvement to another technology or technical field" indicate integration into a practical application. The distinctive tri-part validation gate: image quality validation, refined content extraction validation, and layout validation, applied after postprocessing based on the predicted template, and used to classify a document for a straight through processing or manual annotation, materially improves the technical field of automated document classification and extraction. As disclosed, the pipeline reduces human-in-the-loop rekeying, enhances accuracy by template-conditioned postprocessing, and prevents mis-routing by enforcing layout conformity and quality thresholds before any straight through processing decision; where validation fails, the system classifies the document for manual annotation and optionally learns from the correction via reinforcement learning to improve future performance. Those operations and benefits are technical and computational, not organizational or business-centric.
The Applicant's as-filed specification explains that a straight through processing "aims to streamline and expedite the flow of information...without human intervention in the loop," and that the invention addresses the technical deficiency of conventional systems that are "inefficient in determining if the document is a suitable candidate for an automation process," by predicting a template, computing a confidence score, postprocessing, and then validating along three technical axes before classifying for a straight through processing; otherwise classifying for manual annotation and feeding corrections to learning components. This ordered dependency and classification semantics constitute a specific application of machine learning to images and templates with structured validations, and therefore integrate any abstract idea into a practical application under Step 2A, Prong Two.
In view of the above, the Applicant respectfully submits that amended independent claim 1 integrates the purported judicial exception into a practical application under Prong Two of Step 2A.
The examiner disagrees. First the examiner believes performing a “tri part validation” is a mentally process. Humans can mentally validate different aspects of a document. Humans are also capable of performing ordered processes and classification. The examiner notes that automating mentally processes does not necessarily render them not abstract. The examiner disagrees there is a specific machine learning claimed. Rather the claim states an abstract idea with instruction to perform it with generic “machine learning”. For these reasons the examiner disagrees with applicantss arguments and believes claim 1 and 8 are not integrated into an abstract idea.
Applicant argues with respect to step 2b:
Notwithstanding the above remarks under Step 2A, assuming arguendo, that previously presented independent claim 1 is directed to an abstract idea/judicial exception as the Office Action contends, the Applicant respectfully submits that amended independent claim 1 amounts to "significantly more" than an abstract idea.
Reconsideration and withdrawal of the rejection of independent claim 1 under 35 U.S.C. § 101 is requested in view of the decision of the Court of Appeals for the Federal Circuit regarding subject matter eligibility as acknowledged by the USPTO in its Memorandum issued on November 2, 2016, addressing "Recent Subject Matter Eligibility Decisions (BASCOM Global Internet Services v. AT&T Mobility LLC)." The USPTO, in the Memorandum, instructs "[t]he BASCOM court agreed that the additional elements were generic computer, network, and Internet components that did not amount to significantly more when considered individually, but explained that the district court erred by failing to recognize that when combined, an inventive concept may be found in the non-conventional and non-generic arrangement of the additional elements ... (note that the term 'inventive concept' is often used by the courts to describe additional element(s) that amount to significantly more than a judicial exception)"
(emphasis in original). The USPTO, in the Memorandum, further describes "[i]n Step 2B of the USPTO's SME guidance, examiners should consider the additional elements in combination, as well as individually, when determining whether a claim as a whole amounts to significantly more, as this may be found in the non-conventional and non-generic arrangement of known, conventional elements," (emphasis added).
the Applicant submits that, under Step 2B, the combination of elements of amended independent claim 1 yields a non-conventional and non-generic arrangement: an ML-based pipeline that selects a predicted template with highest matching probability, determines a confidence score between input image and predicted template, postprocesses the extracted content based on the predicted template only if the confidence score exceeds a predefined threshold, then applies a tri-part validation (image quality, refined content extraction, and layout) and classifies for a straight through processing only if each validation is successful, else classifies for manual annotation with optional learning from corrections. The Applicant's as- filed specification details concrete implementations of template prediction (convolutional embeddings, sigmoid similarity), postprocessing rules, image quality scoring, layout field alignment checks, and bounds validation, none of which are routine "data gathering" or mere "display" steps. Amended independent claim l's arrangement therefore amounts to significantly more because it constrains when and how data flows are postprocessed, validated, and classified in a manner that produces technological improvements in accuracy and automation of document extraction.
Therefore, for the reasons presented above, amended independent claim 1 is not directed to an abstract idea under Step 2A, Prong One. In the alternative, said claim integrates any alleged exception into a practical application under Step 2A, Prong Two by reciting a specific technical improvement in automated document extraction and routing via template-conditioned postprocessing and tri-part validation gates that control straight through processing/manual annotation classification. In the further alternative, said claim amounts to significantly more under Step 2B because the non-conventional arrangement of ML-based template selection, confidence gating, postprocessing, tri-part validation, and bifurcated classification yields technical benefits in accuracy, automation, and data integrity.
The examiner notes that “significantly more” refers to additional elements. The elements “pipeline that selects a predicted template with highest matching probability, determines a confidence score between input image and predicted template, postprocesses the extracted content based on the predicted template only if the confidence score exceeds a predefined threshold, then applies a tri-part validation (image quality, refined content extraction, and layout) and classifies for a straight through processing only if each validation is successful, else classifies for manual annotation with optional learning from corrections” Are all mental concepts and therefore cannot be significantly more. The examiner note that while machine learning is not abstract by itself this merely limits the abstract to the field of machine learning and is well known.
Applicant's arguments filed 1/23/2026 with respect to U.S.C. 103 have been fully considered but they are not persuasive.
Applicant argues:
Even assuming familiarity with confidence thresholds and template selection, the Examiner has not identified, within Fujimoto or any other cited reference, a teaching or suggestion to re-architect the pipeline into the claimed postprocessing-then-validation triad and to make straight through processing classification contingent on the collective success of all three validations. The Office Action's reliance on Fujimoto's confidence audit to meet "validation" is an unsupported equivalence that does not supply the specific claim-required validations nor the ordering and dependency that the claim imposes. Under KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007), an obviousness rationale must identify a reason to make the particular combination with a reasonable expectation of success, and cannot rely on hindsight reconstruction. Here, no reasoned motivation is given to refashion Fujimoto's audit-and-manual-keying flow into a template-conditioned postprocessing followed by three validations that conjunctively gate a straight through processing classification; nor does Fujimoto teach or suggest that such a tri-validation framework is used to formally classify documents for a straight through processing.
In addition, the Office Action's rejection does not make particularized findings showing where Fujimoto teaches "validating the extracted content based on postprocessing of the extracted content," "wherein the validation comprises an image quality validation, a refined content extraction validation, and a layout validation," followed by "classifying the document for a straight through processing" only upon successful validation against each validation, as required under In re Kotzab, 217 F.3d 1365 (Fed. Cir. 2000) (requiring specific findings for the proposed combination) and In re Lee, 277 F.3d 1338 (Fed. Cir. 2002) (prohibiting conclusory assertions without evidentiary support). The cited portions of Fujimoto instead discuss audit score generation (e.g., Figures 8A-8D, audit engine) and manual labeler correction when confidence is low, which are conceptually distinct from the claim's tri-validation gate and straight through processing classification tied to that gate.
The examiner notes that Fujimoto discloses postprocessing the extracted content based on the predicted template (see paragraph 137 and 139 note template may be used to assist in detecting the layout of a document “the template image and the associated template with template area coordinates advantageously allow for the extraction of data from an image and to accurately associate the extracted data with a field and an individual”) and validating the extracted content based on postprocessing of the extracted content, wherein the validation comprises, a refined content extraction validation (see paragraph 114 confidences that predicted text is correct and 104 note that the confidence is sued to determined if the image needs to be manually keyed note that images with low confidence scores are manually edited ) As disclosed in prior the rejection and below and classifying the document for a straight through processing, only upon successful validation. The additional elements are disclosed in Rastogi US 2022/0284215 in further view of Pizzocchero US 2023/0281821. Applicant is arguing the references separately here claiming that Fujimoto does not disclose the entirety of the features. Rastogi and Pizzocchero disclose the additional validations which must be passed.
Applicant further argues:
Further, the Applicant submits that Rastogi and Pizzocchero, either alone or in
combination, are neither asserted to nor do they disclose, teach, or suggest the above-mentioned feature as recited in independent claim 1. In particular, Rastogi's approach differs as it just does the layout match based on exact template match which triggers manual annotation below the threshold or sends the document for extraction above the threshold. This approach is based on structure and semantic steps of documents using text, needs domain specific knowledge and background knowledge, unlike the claimed approach which is based on visual features and trained using one-shot learning. Further, Pizzocchero determines document authenticity based on an authentication score derived from image-based region-of-interest reconstruction and template matching, but it does not disclose, teach, or suggest classifying a document for a straight through processing contingent on a tri-part validation (image quality, refined content extraction, and layout validation) applied after postprocessing of extracted content.
The examiner disagrees. First elements like one shot learning are not claimed in claim 1 or 8. Further. Applicant is considering only what the references teach separately rather that what they teach combined. The examiner notes that combined the references teach three types of validation with and therefore teach applicants tri-validation. Fujimoto is relied upon to teach the classification based upon validation. Applicant’s argument none of the references teach all the validation type is unpersuasive because these references teach the elements in combination
Applicant further argues
Furthermore, the Applicant addresses the patentability of the newly added feature reciting "classifying the document for a manual annotation, different from the straight through processing if the extracted content is unsuccessfully validated corresponding to each of the image quality validation, the refined content extraction validation, and the layout validation" below. This amendment formalizes the complementary classification outcome when the claimed tri-part validation fails, and, critically, ties manual annotation classification to the outcome of the same validation framework that follows postprocessing based on the predicted template.
Fujimoto does not anticipate, teach, or suggest the newly added feature. In particular, Fujimoto's routing for human intervention is triggered by low confidence scores predicted by audit engines; it is not a classification driven by the unsuccessful outcome of the specific three validations after template-conditioned postprocessing, as claimed. Fujimoto's "manual keying" is confidence-driven, not validation-conditioned, and its audit signal does not correspond to the claim's explicit image quality validation, refined content extraction validation, and layout validation. The present claim thus introduces a distinct decision logic, namely, a validation-conditioned classification for manual annotation that mirrors the straight through processing gate but operates on unsuccessful validation, which Fujimoto neither discloses nor suggests.
Again the applicant is arguing the references separately the references combined teach three types of validation. One of ordinary skill could have validated using the three types of validation disclosed in the references in combination. Combined the reads on the tri-validation element. Just because all three element are not found in a single reference does not mean they can not be taught by the combination.
Even if secondary references such as generic "layout similarity" or "image quality metrics" were considered, the claim remains non-obvious because it requires (i) postprocessing the extracted content based on the predicted template, and then (ii) applying a three-part validation where all must pass to reach a straight through processing classification and any failure results in manual annotation classification; this specific ordered dependency and conjunctive/disjunctive gate is not found in Fujimoto and has not been shown obvious by the Examiner. Under MPEP §2143, a proper rationale must identify a reason that a POSITA would have re-engineered Fujimoto's confidence-audit flow into this tri-validation-controlled classification regime; no such reason is provided.
The examiner notes that Fujimoto discloses validating to determine if a document needs manual processing or not by one metric, Pizzocchero and Rastogi disclose additional metrics for validating. It would have been easy for one of the ordinary skill in the art to additionally perform the metrics to validate the document. In this simple combination additional metrics are used to validate. This does not require a reengineering but a simple addition of additional metrics.
Applicant further argues:
Further, the Applicant emphasizes that the claim's "refined content extraction validation" expressly occurs based on postprocessing of the extracted content using the predicted template, and is thus a second-stage check distinct from initial extraction. The claim's "layout validation" is a structural compliance check of the similarity/alignment between the predicted template and the input image, and "image quality validation" operates as a gate ensuring that the input image quality is sufficient for a straight through processing. These three validations are separately named and jointly required, and they control classification outcomes: a straight through processing only if each validation is successful; manual annotation if validation is unsuccessful. Fujimoto does not disclose this validation framework, nor does it present a classification step for a straight through processing tied to such validations
The examiner notes that the Fujimoto does disclose postprocessing with the template to refine the extraction of the text and refined content extraction validation validates the extracted text meeting which are separate features. Other references besides Fujimoto are relied upon to show the other two validation features. Second no such “only if each validation is successful” is claimed a successfully validation only “corresponding to each of the image quality validation, the refined content extraction validation, and the layout validation” There is no requirement that all must pass. Even if there was the examiner believes it obvious in view of the cited references. Each validation individually is used to fail a validation so it follows if any one fails the validation would fail.
In summary, the Applicant submits that Fujimoto's confidence-based audit and manual keying do not teach or suggest the claim's postprocessing-driven, tri-part validation gate and the resulting classification for a straight through processing only upon successful validation corresponding to each validation, nor the complementary classification for manual annotation upon unsuccessful validation. The Office Action's reliance on Fujimoto's paragraph [0104] does not satisfy the claim as a whole. Since Rastogi and Pizzocchero, either alone or in combination, are not asserted to disclose, teach, or suggest the feature "classifying the document...," the Applicant respectfully submits Rastogi and Pizzocchero, either alone or in combination, do not disclose, teach, or suggest the newly added feature, "classifying the document for a manual annotation, different from the straight through processing if the extracted content is unsuccessfully validated corresponding to each of the image quality validation, the refined content extraction validation, and the layout validation" as recited in amended independent claim 1. In particular, low-confidence documents are classified for manual review while others are passed to downstream system without manual intervention unlike sending document for extraction or for manual annotation for training as cited in Rastogi's approach. Further, Pizzocchero lacks any teaching of a complementary classification for manual annotation triggered by failure of all three validations; its fallback is retrying or raising an exception when authentication is inconclusive, not routing the document for manual annotation based on unsuccessful multi-dimensional content validation.
The examiner notes again the applicant is arguing the references separately instead of what they teach combined. Combined they teach validating by three different validation metrics. Fujimoto discloses classification based on validation. For the reasons articulated below It would have been obvious to combine the three-validation metrics with Fujimoto. Such a combination would require validation of all three metrices to confirm the document does not require manual intervention
The arguments for claims 3,4, 7,10,11 and 14 rely on applicants’ argument for claims 1 and 8 and are therefore also unpersuasive.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claim 1, 3, 4, 7, 8, 10, 11,and 14 rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Re claim 1
The limitation of receiving, an input image of the document, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, receiving in the context of this claim encompasses looking at the image to mentally acquire it.
The limitation of performing a content extraction from the input image using a document extractor wherein the content extraction indicates extracting data from the input image into at least one of a constant field, a discrete field, and a variable field respectively, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, performing content extraction in the context of this claim encompasses mentally extracting the content.
The limitation of selecting a template from a template dataset using a prediction model, based on the input image, wherein the predicted template has highest matching probability with the input image, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, selection in the context of this claim encompasses mentally performing the selecting.
The limitation of determining a confidence score by comparing the input image and the predicted template, wherein the confidence score indicates similarity between the predicted template and the input image, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, determining in the context of this claim encompasses mentally performing the determination.
The limitation of postprocessing the extracted content based on the predicted template upon determining that the confidence score is more than a pre-defined threshold confidence score, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, postprocessing in the context of this claim encompasses mentally performing the processing.
The limitation of validating the extracted content based on postprocessing of the extracted content, wherein the validation comprises an image quality validation, a refined content extraction validation, and a layout validation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, validation in the context of this claim encompasses mentally performing the validation.
The limitation of classifying the document for a manual annotation, different from the straight through processing if the extracted content is unsuccessfully validated corresponding to each of the image quality validation, the refined content extraction validation, and the layout validation., as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, classifying in the context of this claim encompasses mentally performing the classification.
The limitation of classifying the document for the straight through processing if the extracted content is successfully validated corresponding to each of the image quality validation, the refined content extraction validation, and the layout validation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, classifying in the context of this claim encompasses mentally performing the classification.
If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. In particular, the claim
only recites additional elements – a storage device machine learning (ML) model. The storage device is recited at a high-level of generality (i.e., as a generic storage for storing data) such that it amounts no more than mere instructions to apply the exception using a generic computer component. Using machine learning model to perform the abstract idea merely limits the claim to a generic machine learning environment. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The claim is directed to an abstract idea.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, The storage device is recited at a high-level of generality (i.e., as a generic storage for storing data) such that it amounts no more than mere instructions to apply the exception using a generic computer component. Using machine learning model to perform the abstract idea merely limits the claim to a generic machine learning environment. Mere instructions to apply an exception using a generic computer component and limiting the claim to a generic machine learning environment cannot provide an inventive concept. The claim is not patent eligible
Re claim 3 The limitation of wherein prior to postprocessing the extracted content: classifying the document for the manual annotation, different from the straight through processing if the confidence score is less than the pre-defined threshold confidence score., as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, classifying in the context of this claim encompasses mentally performing the classifying.
The analysis with respect to integration into an abstract idea and significantly more with respect to this claim is not significantly different than the claim from which it depends.
Re claim 4 this claim recites the same abstract idea as the abstract idea of claim 1
This judicial exception is not integrated into a practical application. In particular, the claim
only recites additional elements – a storage device, a machine learning (ML) model and training the machine learning model using a few shot learning technique. The storage device is recited at a high-level of generality (i.e., as a generic storage for storing data) such that it amounts no more than mere instructions to apply the exception using a generic computer component. Using machine learning model to perform the abstract idea merely limits the claim to a generic machine learning environment and training using a few shot learning technique is a well-known machine learning method See Beaver US 2020/0320134 paragraph 41. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The claim is directed to an abstract idea.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, The storage device is recited at a high-level of generality (i.e., as a generic storage for storing data) such that it amounts no more than mere instructions to apply the exception using a generic computer component. Using machine learning model to perform the abstract idea merely limits the claim to a generic machine learning environment and training using a few shot learning technique is a well known machine learning method See Beaver US 2020/0320134 paragraph 41. Mere instructions to apply an exception using a generic computer component and limiting the claim to a generic machine learning environment using a well known machine learning algorithm cannot provide an inventive concept. The claim is not patent eligible
Re claim 7 The limitation of wherein prior to postprocessing the extracted content: classifying the document for the manual annotation, different from the straight through processing if the confidence score is less than the pre-defined threshold confidence score., as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, classifying in the context of this claim encompasses mentally performing the classifying.
The analysis with respect to integration into an abstract idea and significantly more with respect to this claim is not significantly different than the claim from which it depends.
Re claim 8
The limitation of receive, an input image of the document, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, receiving in the context of this claim encompasses looking at the image to mentally acquire it.
The limitation of perform a content extraction from the input image using a document extractor wherein the content extraction indicates extracting data from the input image into at least one of a constant field, a discrete field, and a variable field respectively, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, performing content extraction in the context of this claim encompasses mentally extracting the content.
The limitation of select a template from a template dataset using a prediction model, based on the input image, wherein the predicted template has highest matching probability with the input image, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, selection in the context of this claim encompasses mentally performing the selecting.
The limitation of determine a confidence score by comparing the input image and the predicted template, wherein the confidence score indicates similarity between the predicted template and the input image, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, determining in the context of this claim encompasses mentally performing the determination.
The limitation of postprocess the extracted content based on the predicted template upon determining that the confidence score is more than a pre-defined threshold confidence score, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, postprocessing in the context of this claim encompasses mentally performing the processing.
The limitation of validate the extracted content based on postprocessing of the extracted content, wherein the validation comprises an image quality validation, a refined content extraction validation, and a layout validation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, validation in the context of this claim encompasses mentally performing the validation.
The limitation of classify the document for the straight through processing if the extracted content is successfully validated corresponding to each of the image quality validation, the refined content extraction validation, and the layout validation, as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, classifying in the context of this claim encompasses mentally performing the classification.
If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
This judicial exception is not integrated into a practical application. In particular, the claim
only recites additional elements – a storage device machine learning (ML) model and a processor coupled to memory. The storage device processor and memory are recited at a high-level of generality (i.e., as a generic storage for storing data a generic processor and a generic memory) such that it amounts no more than mere instructions to apply the exception using a generic computer component. Using machine learning model to perform the abstract idea merely limits the claim to a generic machine learning environment. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The claim is directed to an abstract idea.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the storage device processor and memory are recited at a high-level of generality (i.e., as a generic storage for storing data a generic processor and a generic memory such that it amounts no more than mere instructions to apply the exception using a generic computer component. Using machine learning model to perform the abstract idea merely limits the claim to a generic machine learning environment. Mere instructions to apply an exception using a generic computer component and limiting the claim to a generic machine learning environment cannot provide an inventive concept. The claim is not patent eligible
Re claim 10 the limitation of wherein prior to postprocessing the extracted content: classifying the document for the manual annotation, different from the straight through processing if the confidence score is less than the pre-defined threshold confidence score., as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, classifying in the context of this claim encompasses mentally performing the classifying.
The analysis with respect to integration into an abstract idea and significantly more with respect to this claim is not significantly different than the claim from which it depends.
Re claim 11 this claim recites the same abstract idea as the abstract idea of claim 8
This judicial exception is not integrated into a practical application. In particular, the claim
only recites additional elements – a storage device machine learning (ML) model and a processor coupled to memory. The storage device processor and memory are recited at a high-level of generality (i.e., as a generic storage for storing data a generic processor and a generic memory) such that it amounts no more than mere instructions to apply the exception using a generic computer component. Using machine learning model to perform the abstract idea merely limits the claim to a generic machine learning environment and training using a few shot learning technique is a well-known machine learning method See Beaver US 2020/0320134 paragraph 41. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The claim is directed to an abstract idea.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, The storage device processor and memory are recited at a high-level of generality (i.e., as a generic storage for storing data a generic processor and a generic memory) such that it amounts no more than mere instructions to apply the exception using a generic computer component. Using machine learning model to perform the abstract idea merely limits the claim to a generic machine learning environment and training using a few shot learning technique is a well known machine learning method See Beaver US 2020/0320134 paragraph 41. Mere instructions to apply an exception using a generic computer component and limiting the claim to a generic machine learning environment using a well known machine learning algorithm cannot provide an inventive concept. The claim is not patent eligible
Re claim 14 The limitation of wherein prior to postprocessing the extracted content: classifying the document for the manual annotation, different from the straight through processing if the confidence score is less than the pre-defined threshold confidence score., as drafted, is a process that, under its broadest reasonable interpretation, covers performance of the limitation in the mind. For example, classifying in the context of this claim encompasses mentally performing the classifying.
The analysis with respect to integration into an abstract idea and significantly more with respect to this claim is not significantly different than the claim from which it depends.
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.
Claim(s) 1, 3, 7, 8,10, and 14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Fujimoto US 2023/0010202 in view of Rastogi US 2022/0284215 in further view of Pizzocchero US 2023/0281821.
Re claim 1
Fujimoto discloses A method of classifying a document for a straight-through processing (STP) using memory enabled modelling (see paragraph 218 note that LSTM models are used), the method comprising:
receiving, from a storage device, an input image of the document (see paragraph 61 and 100 note that the records may come from a data bases storing images of documents);
performing a content extraction from the input image using a document extractor machine learning (ML) model, wherein the content extraction indicates extracting data from the input (see paragraph 121 note that machine learning may be used to divide the genealogical record to area, also see paragraph 125note that the divide areas my further by identified by type of field )image into at least one of a constant field (see paragraph 124 note that a machine printed text area may be determined such as “father” which is printed in the form and would not change see paragraph 119 ), a discrete field (see paragraph 127 note that a check box type field may be identified), and a variable field respectively (see paragraph 125 and 128 note that hand written area may be determined note that handwritten information would be variable based the information written in the form see paragraph 119);
selecting a template from a template dataset using a prediction ML model, based on the input image, wherein the predicted template has highest matching probability with the input image (see paragraph 156 note that a template may be selected based on a CNN which anlayzied an input image);
determining a confidence score by comparing the input image and the predicted template, wherein the confidence score indicates similarity between the predicted template and the input image (see paragraph 156 note that a similarity (i.e. confidence) is determined between a query image and a template image note that the template with the highest similarity is chosen see paragraph 157);
postprocessing the extracted content based on the predicted template (see paragraph 137 and 139 note templates may be used to assist in detecting the layout of a document “the template image and the associated template with template area coordinates advantageously allow for the extraction of data from an image and to accurately associate the extracted data with a field and an individual”)
validating the extracted content based on postprocessing of the extracted content, wherein the validation comprises, a refined content extraction validation (see paragraph 114 confidences that predicted text is correct and 104 note that the confidence is used to determine if the image needs to be manually keyed note that images with low confidence scores are manually edited)
classifying the document for the straight through processing if the extracted content is successfully validated (see paragraph 104 note that the confidence is used to determine if the image needs to be manually keyed note that only images with low confidence scores are manually edited, while other are not).
classifying the document for a manual annotation, different from the straight through processing if the extracted content is unsuccessfully validated. (see paragraph 104 note that the confidence is used to determine if the image needs to be manually keyed note that only images with low confidence scores are manually edited while other are not)
Fujimoto does not expressly disclose:
Using the predicted template upon determining that the confidence score is more than a pre-defined threshold confidence score; wherein the validation comprises, a image quality validation, and a layout validation;
Rastogi discloses the predicted template upon determining that the confidence score is more than a pre-defined threshold confidence score (see paragraph 12 note that similarity is compared to a confidence threshold if the metric is below a threshold manual annotation is performed also see abstract note that the template is used to create rules for document extraction ); wherein the validation comprises, and a layout validation (see paragraph 12 note that a layout similarity is compared to a confidence threshold if the metric is below a threshold manual annotation is performed this corresponds to a layout similarity ). The motivation to combine is “efficient and accurate for extracting the information from the image of templatized document” see paragraph 83. One of ordinary skill in the art could have easily used the thresholding operation of Rastogi in the invention of Fujimoto to determine whether or not to perform manual annotation based on whether or not the matched template has high confidence. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Rastogi and Fujimoto to reach the aforementioned advantage.
Rastogi and Fujimoto do not expressly disclose wherein the validation comprises, a image quality validation. Pizzochero discloses wherein the validation comprises, a image quality validation (see paragraph 69 note that global quality metrics are used see paragraph 68 note that an image may be of too low of quality to be properly analyzed) The motivation to combine is to “identify suitable images for analysis” (see paragraph 68). One of ordinary skill in the art could have easily used image quality metrics to further verify the images of Rastogi and Fujimoto. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention combine Rastogi Fujimoto and Pizzocchero to reach the aforementioned advantage.
Re claim 3 Rastogi further discloses classifying the document for the manual annotation, if the confidence score is less than the pre-defined threshold confidence score (see paragraph 12 note that similarity is compared to a confidence threshold if the metric is below a threshold manual annotation is performed also see abstract note that the template is used to create rules for document extraction).
Re claim 7 Fujimoto further discloses wherein postprocessing the extracted content comprises: determining data corresponding to the input image at least one of the constant field (see paragraph 124 note that a machine printed text area may be determined such as “father” which is printed in the form and would not change see paragraph 119 ), the discrete field (see paragraph 127 note that a check box type field may be identified), and the variable field respectively (see paragraph 125 and 128 note that hand written area may be determined note that handwritten information would be variable based the information written in the form see paragraph 119); based on the predicted template (see paragraph 137 “The template image with the associated template may include one or more template area coordinates identifying fields of interest in the physical form captured by the template image. As illustrated in FIG. 6B, the divided areas delineate fields within a genealogical record from which information may be extracted.” Note that the various elements extracted by Fujimoto may be processed based on the template data identifying the areas)).
Re claim 8 Fujimoto discloses A system of classifying a document for processing using a memory enabling model (see paragraph 218 note that LSTM models are used), the system comprising: a memory; at least one processor communicably coupled to the memory, the at least one processor (see paragraph 20 note that a processor and a memory is used) is configured to
receive, from a storage device, an input image of the document (see paragraph 61 and 100 notes that the records may come from a data bases storing images of documents);
perform a content extraction from the input image using a document extractor machine learning (ML) model, wherein the content extraction indicates extracting data from the input (see paragraph 121 note that machine learning may be used to divide the genealogical record to area, also see paragraph 125note that the divide areas my further by identified by type of field )image into at least one of a constant field (see paragraph 124 note that a machine printed text area may be determined such as “father” which is printed in the form and would not change see paragraph 119 ), a discrete field (see paragraph 127 note that a check box type field may be identified), and a variable field respectively (see paragraph 125 and 128 note that hand written area may be determined note that handwritten information would be variable based the information written in the form see paragraph 119);
select a template from a template dataset using a prediction ML model, based on the input image, wherein the predicted template has highest matching probability with the input image (see paragraph 156 note that a template may be selected based on a CNN which analyzed an input image);
determine a confidence score by comparing the input image and the predicted template, wherein the confidence score indicates similarity between the predicted template and the input image (see paragraph 156 note that a similarity (i.e. confidence) is determined between a query image and a template image note that the template with the highest similarity is chosen see paragraph 157);
postprocess the extracted content based on the predicted template (see paragraph 137 and 139 note templates may be used to assist in detecting the layout of a document “the template image and the associated template with template area coordinates advantageously allow for the extraction of data from an image and to accurately associate the extracted data with a field and an individual”)
validate the extracted content based on postprocessing of the extracted content, wherein the validation comprises, a refined content extraction validation (see paragraph 114 confidences that predicted text is correct and 104 note that the confidence is used to determine if the image needs to be manually keyed note that images with low confidence scores are manually edited)
classify the document for the straight through processing if the extracted content is successfully validated (see paragraph 104 note that the confidence is used to determine if the image needs to be manually keyed note that images with low confidence scores are manually edited.)
classifying the document for a manual annotation, different from the straight through processing if the extracted content is unsuccessfully validated. (see paragraph 104 note that the confidence is used to determine if the image needs to be manually keyed note that only images with low confidence scores are manually edited while other are not)
Fujimoto does not expressly disclose:
Using the predicted template upon determining that the confidence score is more than a pre-defined threshold confidence score; wherein the validation comprises, a image quality validation, and a layout validation;
Rastogi discloses the predicted template upon determining that the confidence score is more than a pre-defined threshold confidence score (see paragraph 12 note that similarity is compared to a confidence threshold if the metric is below a threshold manual annotation is performed also see abstract note that the template is used to create rules for document extraction ); wherein the validation comprises, and a layout validation (see paragraph 12 note that a layout similarity is compared to a confidence threshold if the metric is below a threshold manual annotation is performed this corresponds to a layout similarity ). The motivation to combine is “efficient and accurate for extracting the information from the image of templatized document” see paragraph 83. One of ordinary skill in the art could have easily used the thresholding operation of Rastogi in the invention of Fujimoto to determine whether or not to perform manual annotation based on whether or not the matched template has high confidence. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Rastogi and Fujimoto to reach the aforementioned advantage.
Rastogi and Fujimoto do not expressly disclose wherein the validation comprises, a image quality validation. Pizzochero discloses wherein the validation comprises, a image quality validation (see paragraph 69 note that global quality metrics are used see paragraph 68 note that an image may be of too low of quality to be properly analyzed) The motivation to combine is to “identify suitable images for analysis” (see paragraph 68). One of ordinary skill in the art could have easily used image quality metrics to further verify the images of Rastogi. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention combine Rastogi Fujimoto and Pizzocchero to reach the aforementioned advantage.
Re claim 10 Rastogi further discloses classifying the document for the manual annotation, different from the straight through processing if the confidence score is less than the pre-defined threshold confidence score (see paragraph 12 note that similarity is compared to a confidence threshold if the metric is below a threshold manual annotation is performed also see abstract note that the template is used to create rules for document extraction).
Re claim 14 Fujimoto further discloses wherein postprocessing the extracted content comprises: determining data corresponding to the input image at least one of the constant field (see paragraph 124 note that a machine printed text area may be determined such as “father” which is printed in the form and would not change see paragraph 119 ), the discrete field (see paragraph 127 note that a check box type field may be identified), and the variable field respectively (see paragraph 125 and 128 note that hand written area may be determined note that handwritten information would be variable based the information written in the form see paragraph 119); based on the predicted template (see paragraph 137 “The template image with the associated template may include one or more template area coordinates identifying fields of interest in the physical form captured by the template image. As illustrated in FIG. 6B, the divided areas delineate fields within a genealogical record from which information may be extracted.” Note that the various elements extracted by Fujimoto may be processed based on the template data identifying the areas)).
Claim(s) 4 and 11 is/are rejected under 35 U.S.C. 103 as being unpatentable over Fujimoto US 2023/0010202 in view of Rastogi US 2022/0284215 in further view of Pizzocchero US 2023/0281821 in further view of Iliadis US 2021/0149931.
Re claim 4 Fujimoto discloses wherein the document extractor ML model is trained the (see paragraph 97 note the machine learning models are trained). Fujimoto Rastogi and Pizzocchero do not expressly disclose training using a few-shot learning technique. Iliadis discloses using a few-shot learning technique training using a few-shot learning technique (see paragraph 55 and 57). The motivation to train the system using a support set (see paragraph 55). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Fujimoto Rastogi and Pizzocchero to reach the aforementioned advantage.
Re claim 11 Fujimoto discloses wherein the document extractor ML model is trained the (see paragraph 97 note the machine learning models are trained). Fujimoto Rastogi and Pizzocchero do not expressly disclose training using a few-shot learning technique. Iliadis discloses using a few-shot learning technique training using a few-shot learning technique (see paragraph 55 and 57). The motivation to train the system using a support set (see paragraph 55). Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine Fujimoto Rastogi and Pizzocchero to reach the aforementioned advantage.
Allowable Subject Matter
Claim 2, 5, 6, 9 12 and 13 objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SEAN T MOTSINGER whose telephone number is (571)270-1237. The examiner can normally be reached 9AM-5PM.
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, Chineyere Wills-Burns can be reached at (571) 272-9752. 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.
/SEAN T MOTSINGER/Primary Examiner, Art Unit 2673
Prosecution Insights