OCR & Document AI Patents: Who Leads, Where the Gaps Are 2026
- Filing growth has flattened, not accelerated. Annual filings moved from 8 in 2017 to a peak of 62 in 2025, but the 2022 midpoint of 38 shows the climb stalled well before the most recent years — and 2026 is still a partial count.
- No single company dominates the field. The top 5 assignees account for 22.2% of all 316 records in scope, and the top 10 for 34.8% — concentration exists but a long tail of single- and few-filing entrants does most of the rest of the work.
- Image recognition classes crowd out language-specific ones. G06V (image/video recognition) touches 94.9% of records, while speech/audio (G10L) and pictorial communication (H04N) each sit under 5% — layout and script handling is claimed far more densely than multimodal or audio-linked extraction.
What this landscape covers
This dataset tracks 316 published records filed between 2015 and mid-2026 that combine optical character recognition or document-understanding claims with layout analysis, handwriting recognition, table extraction, multilingual script handling, or low-quality scan processing, classified under G06V30, G06V10, or G06F40. It spans enterprise document-processing vendors, cloud platforms, and specialist redaction and archiving firms alongside a large number of single-filing entrants.
Filing activity is concentrated in the United States as receiving office, with meaningful volume in China, India, and the EPO and WIPO/PCT routes — a pattern consistent with a technology that is claimed regionally around large enterprise document workflows rather than filed as a single global family.
Filing trend and technology composition
The filing curve and the IPC mix together show a field that grew steadily through the early 2020s, plateaued, and remains anchored to image-based recognition rather than adjacent speech or video classes.
A decade of filings, with growth flattening after 2022
Annual filings rose from 8 in 2017 to a midpoint of 38 in 2022, then continued to a peak of 62 in 2025. Because publication typically lags filing by around 18 months, the 2026 figure of 18 is a partial count and should not be read as a drop-off.
IPC subclass distribution across the 316 records
G06V (image/video recognition) appears in 94.9% of records, with G06K (data recognition and presentation) and G06F (electronic digital data processing) each present in over a third of the corpus. G06N (AI models) sits at 20.9%, showing that a meaningful minority of filings frame their claims around learned models rather than deterministic recognition pipelines. Because records can carry multiple classes, these shares sum to more than 100% of the 316-record total.
Shares are the percentage of the 316 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on Optical Character Recognition and Document AI with Eureka
This page is one run against one query. Ask Eureka your own question about optical character recognition and document ai and every answer comes back with the patent numbers behind it.
Try EurekaMost-cited prior art and a representative recent filing
Cloud-based methods and systems for integrated optical character recognition and redaction
The filing describes a cloud-agnostic redaction container that extracts pages from multiple documents, determines a processing order based on a load-balancing criterion, and runs OCR and redaction on that order to produce redacted output ready for transmission or storage.Filed by Redactable Inc., published 2023-09-07 as US20230281330A1.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20050259866A1 | Low resolution OCR for camera acquired documents | 168 |
| 2 | US20080097984A1 | OCR input to search engine | 136 |
| 3 | US20120099792A1 | Adaptive optical character recognition on a document with distorted characters | 117 |
| 4 | US7092870B1 | System and method for managing a textual archive using semantic units | 111 |
| 5 | US20170351913A1 | Document Field Detection And Parsing | 110 |
| 6 | US20070086655A1 | Unfolded convolution for fast feature extraction | 104 |
| 7 | US7499588B2 | Low resolution OCR for camera acquired documents | 97 |
| 8 | US20050163377A1 | Systems and methods for biometric identification using handwriting recognition | 95 |
| 9 | US7689613B2 | OCR input to search engine | 83 |
| 10 | US20140067631A1 | Systems and Methods for Processing Structured Data from a Document Image | 79 |
Citation counts favour older filings that have had more time to accumulate citations within the searched corpus — treat them as a signal of influence on the field, not of current commercial relevance.
Each row carries its publication number; clicking a row searches Eureka by that number.
Put your own technology through the same analysis
Eureka on the web
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →MCP server & REST API
When it has to run inside your own pipeline.
Patent search, landscape analysis and assignee resolution as MCP tools. Drop them into any agent framework, or call REST directly.
Browse MCP servers →What the numbers mean for a filing decision
Three patterns stand out once the raw counts are put next to each other: where claim density sits, how citation age skews influence, and how thin the concentration at the top actually is.
The top of the field is thinner than it looks
22.2% of all 316 records in scope belong to the five most active assignees, rising to 34.8% for the top 10. That leaves roughly two-thirds of the corpus spread across a long tail of companies filing once or a handful of times — a structure closer to an open field than a locked-down one.
Recognition claims dominate; multimodal is thin
Nearly every record in scope carries a G06V classification, and over a third also carry G06K or G06F. By contrast, G10L (speech/audio) and H04N (pictorial communication) each sit under 5% of records, suggesting claim space linking OCR output to audio or video pipelines remains comparatively open.
Growth plateaued after the early-2020s climb
Filings grew from 8 in 2017 to 38 at the 2022 midpoint, then to a peak of 62 in 2025. The flat trajectory between 2022 and the peak, rather than a fresh acceleration, suggests the core recognition techniques are now well-explored and later filings are refining rather than opening new ground.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to optical character recognition and document ai, with the prior art for and against each one.
Who is filing, and where activity has cooled
The ranked leaders span enterprise software, document-management specialists, and financial and cloud infrastructure firms, but recent-year momentum data shows several previously active filers with no output in the latest year.
A single leader, then a fast drop-off
The leading assignee holds 22 records, well ahead of the fifth-place assignee at 9 and the tenth-place assignee at 7. That gap between first and fifth place is the steepest part of the ranking; from tenth place down, the field flattens into single- and low-digit filers.
Several established filers show zero recent activity
Multiple assignees that built meaningful portfolios earlier in the period, including large enterprise software and financial-services filers, show zero filings in the latest tracked year. This does not necessarily mean exit from the space — recent filings may simply not have published yet given typical filing-to-publication lag.
Co-filing is rare and concentrated in a few corporate families
Only five co-assignee pairs appear across the corpus, and the strongest pair links two entities within the same corporate group. This points to a field where most patenting is done independently rather than through joint ventures or cross-licensing arrangements between separate companies.
| Assignee | Recent year | YoY |
|---|---|---|
| US BANK NATIONAL ASSOCIATION | 0 | -100% |
| Microsoft Corporation | 0 | — |
| International Business Machines Corporation | 0 | — |
| Kyocera Document Solutions Inc. | 0 | — |
| SAP SE | 0 | — |
| Sony Group Corporation | 0 | — |
| Zoom Communications Inc. | 0 | — |
| REDACTABLE INC | 0 | — |
Where to take this analysis
The dataset points to specific next steps depending on whether the goal is freedom-to-operate, portfolio strategy, or spotting an open filing angle.
Check claim scope against the most-cited prior art
The five highest-cited records set boundaries around low-resolution OCR, adaptive recognition of distorted characters, and search-engine-linked OCR input that any new filing in those areas needs to map against before drafting claims.
Explore claim scope in EurekaModel the white space in under-5%-share classes
Speech/audio and pictorial-communication classes carry the lowest overlap with the core G06V recognition claims, which is where a first-mover claim is least likely to collide with dense prior art.
Map white space in EurekaTrack assignees with zero recent publications
Several previously active filers show no output in the latest tracked year — worth monitoring for either a pause in R&D investment or a publication lag that will surface new filings shortly.
Set up monitoring in EurekaCommon questions about OCR and Document AI patents
Across the 316 records in this dataset, the leading assignee holds 22 records, noticeably ahead of the fifth-ranked assignee at 9 and the tenth-ranked assignee at 7. The field is not dominated by one company: the top 5 assignees together account for only 22.2% of all records, and the top 10 for 34.8%, leaving the majority of filings spread across a long tail of companies with just a handful of filings each. Anyone benchmarking a competitor's position should look at both the raw rank and this concentration figure, since a top-5 placement still leaves most of the field unaccounted for.
Filings grew substantially from 8 in 2017 to 38 at the 2022 midpoint and on to a peak of 62 in 2025, but the growth curve flattened rather than accelerated after 2022. The 2026 figure of 18 looks like a sharp drop, but publication typically lags filing by around 18 months, so the most recent year is always undercounted and should not be read as a genuine decline yet. The overall pattern is one of a field that expanded quickly in the early 2020s and has since plateaued rather than continuing to accelerate.
G06V, covering image and video recognition, appears in 94.9% of the 316 records in scope, making it the dominant classification by far. G06K (data recognition and presentation) and G06F (electronic digital data processing) each appear in roughly a third of records, while G06N (AI-model-based computing) appears in 20.9%, reflecting a meaningful but minority share of filings built around learned models rather than deterministic pipelines. Classes tied to speech, audio, and video communication — G10L and H04N — each sit under 5%, marking them as comparatively open relative to the core recognition classes.
The clearest under-claimed territory sits at the intersection of OCR with classes that have low overlap in this dataset: speech/audio analysis (G10L, 4.1% of records) and pictorial communication tied to video (H04N, 4.4% of records). Cross-script table extraction and low-quality scan restoration specifically for handwriting also show thinner claim density than the core layout-analysis and text-detection claims that dominate the G06V-heavy portion of the corpus. A first claim in these areas would need to tie recognition output directly to an audio, video, or business-process pipeline rather than treating OCR as a standalone image-processing step.
The highest-cited record in the dataset addresses low-resolution OCR for camera-acquired documents, followed closely by OCR input feeding a search engine, adaptive recognition of distorted characters, semantic-unit management of a textual archive, and document field detection and parsing. These citation counts reflect influence within the searched corpus built up over time, so older filings are naturally favoured — they are a signal of foundational status, not necessarily of current commercial relevance. Anyone drafting new claims in low-quality scan handling or field-level document parsing should review these records closely, since they define much of the boundary that later filings had to work around.
Research Optical Character Recognition and Document AI in depth with Eureka
Go past this page: query the whole optical character recognition and document ai corpus yourself, in your own scope.
Every answer comes back with patent numbers you can open.
Disclaimer. This page is generated from Patsnap Eureka data drawn from a limited snapshot of global patent and scientific-literature records, and is provided for general information and reference only.
Patent data carries inherent limitations: recent filings (typically the most recent 18–24 months) are under-counted due to standard publication lag; counts may be reported at either a patent-family or a patent-record basis and are not always directly comparable; classification, applicant-name, and citation data may contain errors, duplicates, or omissions; and the underlying search query defines and constrains the scope shown. As a result, the analysis may be incomplete or inaccurate and may not reflect the full technology landscape.
Nothing on this page constitutes an exhaustive prior-art, novelty, freedom-to-operate, or validity search, nor does it constitute legal, financial, investment, or professional advice, and it should not be relied upon as such. Any patent, commercial, or strategic decision should be verified independently and reviewed with qualified patent, legal, and domain professionals. Patsnap makes no warranties, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of the information presented.
Machine translation. Assignee and organisation names originally recorded in Chinese, Japanese or Korean have been rendered into English by an AI translation step so that the tables stay readable. These renderings are best-effort and may not match a company’s registered English name; the original name is what the underlying patent record carries, and it is what any Eureka query launched from this page uses.