US 6104834

I'll verify the patent details against live sources, including any recent litigation activity.

Summary: US Patent 6,104,834

Title: Matching CCITT compressed document images

Patent number: 6,104,834 (granted from Application No. 08/695,825)

Assignee (as recorded): Ricoh Company Limited (Japan) and Ricoh Corporation (US). Google Patents also lists "Ricoh Americas Corp" among current assignees; USPTO.report lists Ricoh Company Limited and Ricoh Corporation.

Inventor: Jonathan J. Hull (Cupertino, CA) — sole named inventor.

Filing date: August 1, 1996 (Application 08/695,825; also the priority date per Google Patents).

Issue/publication date: August 15, 2000.

Other prosecution details: Primary Examiner Bhavesh Mehta; Assistant Examiner Kanji Patel; attorney/agent Townsend and Townsend and Crew LLP. A continuation (Ser. No. 09/604,783) issued as US 6,546,136 B1 on April 8, 2003. A Japanese counterpart, JP 3766186 B2, lists the same 1996 priority.

Legal status: Expired — Lifetime (maintenance fees paid at years 4, 8, and 12; Google Patents shows anticipated expiration 2016-08-01).

Abstract (verbatim):
"A fast, memory efficient, and accurate document image matching system is disclosed. Document image matching is based on identifying anchor points of characters in the document. The document matching process includes a feature extraction step where anchor points, e.g., points representing approximate locations of characters, are identified as features for matching. In a particularly efficient implementation, the anchor points are 'pass codes' in a two-dimensionally encoded representation of a document image."

Plain-language overview of the independent claims

The patent has 38 total claims with six independent claims — three method claims (1, 28, 36), two computer-program-product claims (11, 32), and one system claim (21). All share the same core pipeline.

Claim 1 — Method (broadest), anchor points + Hausdorff:
Detect changes in pixel values between successive pixels in an input document image; record the locations of those changes as "anchor points"; compare that anchor-point set against previously stored anchor-point sets for candidate documents by computing a Hausdorff distance to each; identify as the best match the candidate with the smallest Hausdorff distance.

Claim 11 — Computer program product:
Same steps as claim 1, recast as executable code plus a computer-readable storage medium; again requires computation of a Hausdorff distance and selection of the smallest-distance candidate.

Claim 21 — Computer system:
A storage device plus a processing system that extracts pass-code locations from a document image (a pass code meaning a run of consecutive black or white pixels on one line has no corresponding run on an adjacent line), compares them to stored pass-code sets using a Hausdorff distance, and identifies the smallest-distance candidate as the match.

Claim 28 — Method on a compressed representation:
Accepts an already-compressed representation of the document that contains pass codes; extracts pass-code locations as the anchor points; compares to stored anchor-point sets via Hausdorff distance; selects the smallest-distance candidate.

Claim 32 — Computer program product:
Substantively the same as claim 28 (compressed input with pass codes → extract locations → Hausdorff compare → smallest distance), expressed as code on a storage medium.

Claim 36 — Method with a 2-D representation:
Detects pixel-value changes, determines their locations, and captures them as a two-dimensional (2-D) representation of extracted features; compares that 2-D representation to previously obtained 2-D representations using a Hausdorff distance; selects the smallest-distance candidate.

Key dependent claims (scope additions):

• Claims 2, 12, 22, 37 — implement the anchor points as pass codes extracted from a compressed representation (this is the "preferred embodiment" hook in independent claims 1, 11, 21 and 36).
• Claims 3/4, 13/14, 23/24, 29/30, 33/34, 38 — specify the compression standard as CCITT Group III and/or Group IV.
• Claims 5/6, 15/16, 25/26 — locate a pass code horizontally by accumulating pixels represented by preceding codewords and vertically by counting end-of-line (EOL) indicators.
• Claim 7/17/27 — anchor-point locations found in a single pass through the image (a stated speed/memory advantage).
• Claim 8/18 — pass codes determined by detecting changing elements.
• Claims 9/10/19/20/31/35 — anchor points represent approximate character locations / anchors of characters.
• Claim 37 — the 2-D representation includes the pass-code locations.

Implementation context disclosed in the specification: the matching may run on a general-purpose computer or inside a digital copier/facsimile machine, where the CCITT-compressed bitstream is already available, avoiding decompression. The specification reports a test on 800 University of Washington CD-ROM document images using one-inch-square patches, with 100% self-match accuracy and 92.5% accuracy in identifying photocopies.

Litigation check (USPTO / CAFC 2026 dockets)

I searched for CAFC 2026 docket activity naming U.S. Patent 6,104,834 and found none. Things worth flagging so you don't get a false positive:

• The Federal Circuit's April 23, 2026 Centripetal Networks v. Keysight Technologies decisions (Nos. 24-1406, 24-1416) concern the '370 and '917 patents — not 6,104,834.
• The 2026 Magnolia Medical Technologies v. Kurin appeal references a "'483 patent" — that is a different patent (a Magnolia blood-diversion patent), not U.S. 6,104,834. The "'483" label refers to a numeric suffix, not this patent number.
• Given the patent expired in 2016, active infringement litigation is unlikely; the 2026 hits returned are unrelated patents.

Confidence / uncertainty notes:

• Patent bibliographic data, abstract, and claim text above are drawn from the authoritative full text plus corroborating Google Patents, FreePatentsOnline, and USPTO.report pages; these agree on the core facts.
• Minor third-party database discrepancies: Unified Patents lists priority/grant dates of 1996-07-31 and 2000-08-14 and an expiration of 2016-07-31, whereas Google Patents and USPTO.report list 1996-08-01 and 2000-08-15. I have not independently confirmed which is the operative USPTO-record date.
• Two apparent typographical anomalies in the granted claim text exist in the official copy: claim 12 depends "of claim 7" (a method claim) rather than claim 11, and claim 27 recites "pass word locations" rather than "pass code locations." I report these literally rather than correcting them.
• I found no 2026 CAFC docket entries for this patent. If you need a definitive docket-negative, that should be confirmed directly against the CAFC/PACER docket system and the USPTO Patent Trial and Appeal Board record, which my search did not query directly.

I'll research litigation involving US patent 6104834. Let me search multiple sources.

Let me check the litigation-specific sources.

Let me check a few more targeted sources including CourtListener and Docket Alarm.

Litigation Search Results: U.S. Patent No. 6,104,834

Bottom line: I found no evidence of any litigation (or PTAB/AIA challenge) involving U.S. Patent No. 6,104,834.

What I searched and found

I searched the Unified Patents patent/litigation portal, CourtListener/Justia RECAP documents, and general web sources for the specific number 6,104,834 (and US6104834A).

• Unified Patents portal — The dedicated page for this patent (https://portal.unifiedpatents.com/patents/patent/US-6104834-A) lists bibliographic data, family members (3), and "Referenced By (123)," but shows no litigation entries and no "Opposition Documents" (i.e., no IPR/PGR filings) associated with the patent.
• No district-court, ITC, or Federal Circuit records surfaced tying this specific patent number to any case caption, docket, or party (other than Ricoh as the named assignee).

Important false-positive warning

My searches returned numerous cases citing an "'834 patent" — but each of these is a different patent number, not 6,104,834. I am explicitly excluding them:

Per your instruction, I did not conflate these with 6,104,834.

Context on the patent itself (from the authoritative patent text)

• Patent: U.S. 6,104,834, "Matching CCITT compressed document images"
• Inventor: Jonathan J. Hull
• Assignee: Ricoh Company, Ltd. / Ricoh Americas Corp. (assignment recorded 1999-03-16; REEL/FRAME 009844/0682, effective 1999-03-10)
• Filing/priority: 1996-08-01; granted 2000-08-15
• Legal status: Expired – Lifetime (anticipated expiration 2016-08-01)
• Continuation: US 6,546,136 B1 (from App. 09/604,783, a continuation)

A patent that (a) expired in 2016 and (b) shows no litigation records in the queried databases has plausibly never been asserted.

Limitations / caveats

I could not complete a definitive PACER docket-name sweep before exhausting my search steps, and PACER full-text docket searches are not freely queryable via general web search. Therefore, while the available litigation databases show nothing, I cannot rule out a low-profile suit that never appeared in the aggregators, nor a Japanese or other foreign proceeding on the sibling members (JP 3766186 B2 / JPH1093817A). If you need certainty, the authoritative next step is a PACER "Case Locator" search and a Lex Machina/Docket Navigator query on the exact patent number 6,104,834.

If no litigation is known: it appears none is known. Based on the sources available to me, I have found no litigation involving U.S. Patent No. 6,104,834.

I'll verify the structured block's "no PTAB activity" default against live sources before writing the analysis.

Let me run a couple more targeted checks specifically against PTAB dockets and litigation/assertion history.

Proceedings overview

Zero AIA trial proceedings on file. The canonical structured list (USPTO Open Data Portal ingest) returns 0 proceedings — no IPR, no PGR, no CBM — and targeted web searches across PTAB dockets, Unified Patents' patent portal, and Google Patents' litigation/PTAB panels surfaced no AIA trial petition, institution decision, FWD, or appeal involving US6104834 (the '834 patent) or its continuation US6546136B1 (the '136 patent). Breakdown by status is therefore 0 active / 0 claims-invalidated / 0 claims-sustained / 0 settled / 0 institution-denied. Defensive posture: the patent is not "hardened" by surviving IPRs — it has never been tested at the Board at all, and all 38 claims remain presumptively valid but statutorily expired. The dominant fact for a defendant today is not PTAB history but the expiration of the patent on 2016-08-01, which caps exposure to past damages and removes most of the economic incentive that would normally attract an IPR filer.

Proceedings

No proceedings to report. Per the structured PTAB block, the ODP API returned no AIA trial proceedings as of the most recent ingest. Independent verification found nothing to add:

• Google Patents — US6104834A (https://patents.google.com/patent/US6104834/en): no PTAB/IPR tab or trial-proceeding entry appears; the "Cited By" list reflects prosecution and post-issuance citations, not AIA challenges to this patent.
• Unified Patents portal — US-6104834-A (https://portal.unifiedpatents.com/patents/patent/US-6104834-A): entry exists for the patent (assignee Ricoh, priority 1996-07-31/1996-08-01, expiration 2016-07-31), but no IPR/PGR/CBM proceeding is associated.
• PTAB E2E (https://ptacts.uspto.gov/ptacts/): no party search returned a petition against the '834 or '136 patent numbers.

I did not find any proceeding number to report, and I will not invent one. If the ODP ingest is stale and a proceeding exists, it would most likely be a pre-2019 filing (the patent expired in 2016), but no such record surfaced.

Strategic summary

Claim status — all 38 claims are UNTESTED, and the patent is expired. US6104834 issued 2000-08-15 from Application 08/695,825, filed 1996-08-01, with a priority date of 1996-08-01 and an anticipated expiration of 2016-08-01 (confirmed in the Legal Events: "2016-08-01 Anticipated expiration" and current status "Expired - Lifetime"). The claim set spans 38 claims across method, computer-program-product, and computer-system formats — independents are claim 1 (method: detect pixel-value changes → capture anchor points → Hausdorff distance comparison), claim 11 (program product), claim 21 (computer system, pass-code extraction), claim 28 (method, compressed-representation input), claim 32 (program product, compressed representation), and claim 36 (method, 2-D feature representation). None of these has ever been construed or adjudicated by the PTAB, because no AIA trial was ever filed.

Estoppel landscape — § 315(e)(2) is a non-issue. Because no IPR/PGR/CBM was instituted, there is no IPR estoppel barring any invalidity ground. A defendant is free to raise any § 102/§ 103/§ 112 theory in district court or in a reexam, including the 13 references cited on the face of the patent (e.g., US 5,465,353 to Hull, US 5,414,781, US 5,245,676) and the non-patent literature (Huttenlocher/Klanderman/Rucklidge, "Comparing Images Using the Hausdorff Distance," IEEE TPAMI 15(9), Sept. 1993; Rucklidge, Cornell TR94-1454; Spitz, "Skew Determination in CCITT Group 4 Compressed Document Images"). Note that Huttenlocher et al. and Rucklidge are the very references the specification incorporates by reference for the Hausdorff metric — a defendant should consider whether they qualify as prior art against the asserted claims independently of that incorporation. That analysis has never been litigated at the Board.

Pattern signals — no petitioner, no aggregator, no appeal. There is no repeat petitioner, no Unified Patents (or other defensive aggregator) challenge in the chain, and no Federal Circuit appeal of any '834 FWD, because none exists. The citation pattern is one-directional: the '834 patent is cited by dozens of later Ricoh mixed-media-reality patents and by third parties (e.g., US 7,359,904; EP 2,405,381), and the family itself generated a continuation ('136) — consistent with a fertile internal research lineage rather than an assertion-driven litigation target. The absence of IPRs is explained most simply by the patent's age and 2016 expiry: it was never a high-value assertion vehicle during the post-AIA era.

Recommended next steps

1. Treat "no PTAB activity" as the operative fact and pivot to expiration. The '834 patent's term ended 2016-08-01 (20 years from the 1996-08-01 filing; Legal Events confirm "Anticipated expiration" on that date; status "Expired - Lifetime"). Verify this against USPTO Patent Center before relying on it, but if accurate, the only surviving exposure is past damages within the § 286 six-year lookback measured backward from any complaint — and for most modern assertion targets that window has closed entirely. An expired patent cannot support injunctive relief and cannot be infringed prospectively.

2. If you receive a demand letter citing US6104834, check the date of the accused conduct first. If all accused activity postdates 2016-08-01, there is no actionable infringement theory on the '834 patent. Do not treat the absence of IPRs as evidence the claims are strong — it is evidence the patent aged out.

3. Confirm no post-expiry continuation or sibling is being asserted. The family includes US6546136B1 (continuation of 08/695,825, filed 2000-06-26, granted 2003-04-08) and JP3766186B2. Search for whether any continuation, divisional, or reissue in this family carries a later expiry, since that is where a live assertion would have to come from. None of these siblings shows PTAB activity either.

4. If any PTAB proceeding does exist but is absent from the ODP ingest, pull it directly from PTAB E2E (https://ptacts.uspto.gov/ptacts/) by patent number and by party name, and cross-check CourtListener (https://www.courtlistener.com/) for any corresponding CAFC appeal. I found no docket number to hand you.

5. Bottom line for the defense file. There is nothing to quote from an FWD, no canceled claim to point to, and no claim sustained over an IPR challenge. The strongest defensive levers are (a) statutory expiration (2016-08-01), (b) the § 286 damages bar on stale conduct, and (c) a clean slate to run any invalidity theory without § 315(e)(2) estoppel — including the face-of-patent references and the Huttenlocher/Rucklidge Hausdorff-metric literature that the specification itself leans on.

Caveat: this conclusion rests on the ODP structured data being current. I could not independently retrieve a PTAB record for US6104834 via web search; I did not find one, and I am flagging that as a verification gap rather than asserting a negative with certainty.

I'll verify the assignment record against live sources before drafting.

The searches surfaced the Google Patents legal events (1999 assignment, reel 009844/0682). Let me confirm the reel/frame record and check for any post-1999 transfers or change-of-name filings.

Let me check whether any change-of-name or later transfer was recorded, and verify the continuation's record.

Assignment & Ownership Analysis — US 6,104,834

Data-access note: I could retrieve Google Patents' legal-events tab and third-party assignment mirrors (plainSite's USPTO Patent Assignment extracts, Unified Patents, uspto.report, FreePatentsOnline), but I could not open the Assignment Center UI directly to pull the raw reel/frame images or the correspondent-of-record field. Where the correspondent is missing I say so rather than guess. Verification links: USPTO Assignment Center · legacy Assignment Search (search by patent number 6104834).

Inventors

Pattern check — inventor longevity: Hull did not depart the assignee. His own CV (rii.ricoh.com/~hull/vita_hull.pdf) lists US 6,104,834 as publication [14], and he continues as a named Ricoh inventor for well over a decade afterward (e.g., the family around US 6,365,381 and the "mixed media environment" family, Ricoh-assigned, 2004–2013). This is the opposite of the "all inventors depart within 12 months" fire-sale precursor — it is a normal long-tenure corporate-researcher profile. No unusual ownership pattern.

Original assignee

Named on the issued patent: Ricoh Company, Ltd. (Japan) and Ricoh Corporation (U.S./California). The 1999 assignment recites the same pair as assignees, so the two-entity ownership is confirmed by record, not just the face of the patent.

• Primary line of business: Ricoh is a large operating manufacturer — office imaging (MFPs, copiers, printers), facsimile equipment, and document-management software. The '834 spec explicitly contemplates embodiment "in the context of a digital copier or facsimile machine," so the claimed subject matter maps directly onto Ricoh's core product lines (a fax receiver would already have the CCITT bitstream in hand — the spec's stated efficiency argument).
• Did they ship a product embodying the claims? The assignee is unquestionably an operating company in the relevant product category, and the specification is drafted around shipping hardware. I did not locate a specific product SKU identified as practicing the '834 claims, so I won't assert product-level mapping I can't cite.
• Current status: Operating, solvent, not acquired, not in bankruptcy. The U.S. entity's corporate history is a naming story, not a distress story: Ricoh Industries U.S.A., Inc. → Ricoh Corporation → Ricoh Americas Corporation (per Ricoh's own annual-report milestones), and as of April 1, 2016 Ricoh USA, Inc. and Ricoh Americas Corporation were consolidated into Ricoh USA, Inc. Ricoh Company, Ltd. is the Tokyo-listed parent.

Cross-reference: the JP counterpart JP 3766186 B2 is the same 1996 priority family and was likewise a Ricoh-owned, Ricoh-prosecuted filing — no separate owners appear.

Assignment timeline

Two recorded events of any kind touch this chain; only one is a transfer of ownership.

1. 1999-03-10 (executed) / recorded 1999-03-16 — Reel 009844/0682

• Conveyance: Assignment (Assignment of Assignors Interest)
• Assignor: Jonathan J. Hull
• Assignee: Ricoh Corporation (California) and Ricoh Company Limited (Japan)
• Correspondent: Not retrievable from the sources I could open. The only practitioner identified on the face of the patent is Townsend and Townsend and Crew LLP (prosecution attorney/agent of record, per uspto.report) — a mainstream Silicon Valley IP firm, not an NPE recording mill. I flag the correspondent-field gap explicitly rather than infer.
• Context: Original inventor-to-employer assignment — the standard employment paperwork, executed ~7 months after the 1996-08-01 filing.

2. 2007-04-01 (executed) / recorded 2010-05-21 — Reel 24422/0961

• Conveyance: Change of Name (see document for details)
• Assignor: Ricoh Corporation
• Assignee: Ricoh Americas Corporation, 5 Dedrick Place, West Caldwell, NJ 07006
• Correspondent: Not retrievable from the sources I could open.
• Context: Change of name only — no transfer of ownership, no consideration, no third party. Corporate renaming of the U.S. Ricoh entity; recorded as a bulk/multi-patent recording, not a patent-specific filing.

⚠️ Discrepancy to flag (builds on the earlier summary, which noted Ricoh Americas Corp appearing among current assignees): Google Patents' legal-events tab for US 6,104,834 shows only the 1999 assignment plus maintenance-fee payments — it does not list reel 24422/0961. The change-of-name record surfaced via plainSite's USPTO assignment extract (Assignment 24422/961). I could not confirm that this bulk recording enumerates US 6,104,834 individually. Treat the Ricoh Corporation → Ricoh Americas Corporation name change as confirmed as a corporate fact but as unconfirmed at the per-patent reel level for this specific patent.

Family note: Continuation US 6,546,136 B1 (Ser. No. 09/604,783, filed 2000-06-26, issued 2003-04-08) carries the same title and 1996-08-01 priority and is a Ricoh-family application; no separate assignment record for it surfaced in my searches, and I did not reach a definitive per-application check.

No other recorded assignments exist. There is no post-1999 transfer to any third party, LLC, or aggregator. As of the 2016-08-01 expiration the patent remained with the Ricoh entity that took the 1999 assignment (through the 2007/2010 name change).

Timeline diagram

timeline
    title Ownership of US 6104834
    1996 : Application filed by Hull and Ricoh
    1999 : Hull assigns rights to Ricoh
    2000 : Patent issues to Ricoh
    2007 : Ricoh Corporation renamed Ricoh Americas
    2010 : Name change recorded at USPTO
    2016 : Patent expires in Ricoh hands

NPE / troll-pattern signals

Verdict

Insufficient data — but not for want of records: the chain consists of only the original assignment (Reel 009844/0682, executed 1999-03-10, recorded 1999-03-16, Hull → Ricoh Corporation and Ricoh Company Limited), plus a change-of-name-only recording (Reel 24422/0961, executed 2007-04-01, recorded 2010-05-21, Ricoh Corporation → Ricoh Americas Corporation). That is the classic "no post-issuance deal" profile, which in this rubric lands on Insufficient data because there is no assertion history to classify.

Substantively, every NPE signal is negative. This is a garden-variety employee-inventor assignment to a large operating manufacturer that kept the patent for its entire life, paid all three maintenance fees (years 4, 8, 12), and let it expire on 2016-08-01. There is no shell LLC, no repeat recording attorney to trace, no cascade, no fire-sale, and no defensive aggregator on the other end — so neither "NPE" nor "defensive aggregator" is the right box. If you want a single sentence: this patent was never touched by the NPE ecosystem.

One caveat to carry forward: the per-patent correspondence between Reel 24422/0961 and US 6,104,834 is unverified (Google Patents omits it), so if the ownership question is ever outcome-critical, pull Reel 24422/0961 directly from the Assignment Center to confirm it enumerates this patent.

I'll research this systematically — first verifying the patent itself at USPTO, then examining each of the 13 cited references to build the § 102 anticipation analysis.

Excellent — I have strong material on the most relevant references. Let me gather the remaining cited patents and the non-patent literature.

Prior-Art Analysis — U.S. Patent No. 6,104,834

1. USPTO / publication identification (number-specific)

I searched specifically for 6,104,834 and excluded similar numbers.

Sources: Google Patents, USPTO.report, Unified Patents portal, Espacenet.

Dates to hold constant (from the authoritative patent text): the '834 patent's 13 cited U.S. patents span issued dates 1990-07-10 to 1998-01-13. Because the application was filed 1996-08-01, the pre-AIA statute governs. The two legally material reference dates are the § 102(b) one-year bar of 1995-08-01 and the § 102(e) "filed-before" date (each reference's own U.S. filing date).

Discrepancy flag (carried over): Unified Patents reports priority 1996-07-31 and grant 2000-08-14 (expiration 2016-07-31); Google Patents/USPTO.report report 1996-08-01 and 2000-08-15. The operative dates do not change the anticipation conclusions below. I also confirm the two printed-claim anomalies previously flagged — claim 12 depends "of claim 7" (a method claim) and claim 27 recites "pass word locations" — and I leave them uncorrected per the literal-interpretation rule.

2. Legal framework applied

• Pre-AIA 35 U.S.C. § 102 requires that a single reference disclose every limitation of the claim, arranged as in the claim, to anticipate. A reference that discloses only some limitations cannot anticipate; it can only support a § 103 obviousness rejection when combined with other art.
• Consequence for this patent: the independent claims (1, 11, 21, 28, 32, 36) each require (i) an anchor/pass-code extraction step and (ii) a Hausdorff-distance comparison that selects the smallest-distance candidate. A reference that supplies only (i) cannot anticipate those claims, and because dependent claims incorporate all limitations of their parents, such a reference cannot anticipate the dependents either.
• Accordingly, the analysis below reports (a) whether any claim as a whole is anticipated, and (b) which claim limitations the reference supplies (the § 103 building blocks).

3. The 13 cited U.S. patents

Dates below are the patent-number / filing / publication dates as printed on the '834 front page.

Notes on the entries

Reference 3 — US 5,245,676 (Xerox). Verified from the full text and claims (Justia/FPO). It discloses scanning a binary image, compressing it (e.g., CCITT G4/G3 two-dimensional), and using pass-code locations as "fiducial points." Claim 5 recites selecting, as topographic features, "the locations of pixel color transition … for a selected one of said scan lines which do not have a corresponding pixel color transition … for an adjacent one of said scan lines before a color transition … on said selected one of said scan lines" — i.e., the artifact the '834 patent calls a pass code. Claim 12 expressly covers "a CCITT compression format that includes pass codes … obtaining coordinates of a pass code." Its EP sibling EP 0 434 415 A3 states the same ("the locations of the fiducial points may be taken as the locations of the pass codes … pass codes indicating a pass of white pixels"). This supplies the pass-code-extraction limitation of claims 2, 21 and 28 (and the CCITT limitation of claims 3/4, 23/24, 29/30). It does not disclose any Hausdorff-distance comparison or "smallest distance" selection, so it cannot anticipate claim 1, 11, 21, 28, 32 or 36 as a whole.

Reference 8 — US 5,414,781 (Xerox) — the closest patent citation. Verified (Google Patents, FPO). It discloses: scanning a document; compressing the data "according to the CCITT Group 4 or Group 3 two-dimensional encoding standard" (claim 13); "identifying pass codes in said compressed data and the locations thereof" (claim 12); determining the distribution of the power of the alignment angles of the locations of said pass codes; and comparing that distribution "with the corresponding distribution characteristic of known logotypes … to determine the closest match and the document classification" (claim 12). It also confirms the physical insight the '834 patent relies on — white passes are "indicative of the bottoms of black structures," and "there is at least one white pass at the bottom of each connected component."

This is materially closer than reference 3 because it performs pass-code location extraction + comparison to stored reference data to find a closest match. It overlaps the pass-code limitations of claims 2–4, 21–24, and 28–30. But its matching metric is an alignment-angle power distribution, not a Hausdorff distance, and it selects a "closest match" on that distribution — not a "smallest Hausdorff distance." It therefore does not anticipate the independent claims as written. It is the strongest single § 103 reference for the pass-code limbs of the claims.

Reference 10 — US 5,465,353 (Ricoh / Hull). Verified (Google Patents, Typeset). Same inventor as the '834 patent. It discloses matching an input document image against a document database using descriptors derived from "interesting points," a descriptor database, accumulator voting, and returning the document with the most votes. It is the "matching document images" ancestor of the '834 work, but it uses hashing/voting, not Hausdorff distance, and never mentions pass codes. It supplies only the general two-stage (extract features → match against database) framework, not any limitation unique to the '834 claims.

Reference 12 — US 5,533,144 (Xerox). Anti-counterfeit detector; issued 1996-07-02, i.e., less than one year before the '834 filing (so § 102(a)/(e), not § 102(b)). Thematically about pattern detection/authentication, not document-image matching; not anticipatory.

Reference 7 — US 5,267,047 (IBM). Verified (Google Patents, EveryPatent). A facsimile-server/image-archiving subsystem; its IOCA Function Set 10 supports "the IBM modified CCITT Modified Read Algorithm and the CCITT T.6 Group 4 Facsimile Coding Scheme." It is background showing that facsimile/archiving pipelines already carry CCITT-compressed images — relevant to the "facsimile receiver" implementation context, not to the matching claims.

References 1–2 (Iterated Systems), 6 (Optigraphics), 9, 11, 13. Fractal/IFS compression (1, 2), tiled multi-resolution image management (6), note-hyperlinking (9), text-selection granularity (11), and automatic summary/hyperlink generation (13). None discloses document-image matching by anchor points, pass codes, or Hausdorff metrics. They appear to have been cited as general image-processing/compression or document-handling background.

Reference 4 (Hitachi), 5 (Sussman) — abstract not retrieved. My search budget was exhausted before I could pull confirmed abstracts for US 5,388,167 and US 5,586,196. On their face (a document image-processing system; a digital document magnifier) neither is directed to pass-code-based Hausdorff matching, but I flag that I did not verify their disclosures line-by-line. If a definitive element-by-element read is required, these two should be pulled and read in full.

4. Non-patent literature (arguably the most relevant prior art)

The front page lists 8 NPL entries (4 unique documents):

1. A. Lawrence Spitz, "Skew Determination in CCITT Group 4 Compressed Document Images," pp. 11–25 (1989). Establishes that features can be extracted directly from a CCITT Group 4 bitstream — the conceptual ancestor of the pass-code idea, and the precursor to Xerox references 3 and 8.
2. Huttenlocher, Klanderman & Rucklidge, "Comparing Images Using the Hausdorff Distance," IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 15, no. 9, Sept. 1993, pp. 850–863. This is the source of the metric recited in every independent claim. The '834 specification expressly incorporates it and adopts its "modified Hausdorff measure" h_k(A,B) = k-th ranked min-distance. It discloses comparing point sets, the k-th-ranked (outlier-robust) variant, and using dilation/correlation to count points within δ — i.e., the substance of the "Hausdorff distance" limitations of claims 1, 11, 21, 28, 32 and 36.
3. Jonathan J. Hull, "Document Image Matching and Retrieval With Multiple Distortion-Invariant Descriptors," IAPR Workshop on Document Analysis Systems, Series in Machine Perception and AI, vol. 14, World Scientific, 1995, pp. 379–396. Same-inventor work on document-image matching with distortion-invariant descriptors (the academic counterpart of reference 10).
4. William Rucklidge, "Efficient Computation of the Minimum Hausdorff Distance for Visual Recognition," TR94-1454, Cornell Univ. Dept. of Computer Science, Sept. 1994, pp. 1–169. Discloses efficient minimum-Hausdorff computation and the translation-search ("trying a range of t values") referenced in the '834 specification.

Key point: the '834 patent's novelty lies in the combination — using CCITT pass codes as the anchor-point set (Spitz / US 5,245,676 / US 5,414,781) and comparing those sets with a (modified) Hausdorff distance (Huttenlocher et al. / Rucklidge). No single reference, patent or NPL, teaches that combination.

5. Bottom-line § 102 findings

1. No cited reference — patent or non-patent — anticipates any of the 38 claims as a whole. Each independent claim (1, 11, 21, 28, 32, 36) requires both a pass-code/anchor extraction step and a Hausdorff-distance comparison selecting the smallest-distance candidate. No single reference discloses both.
2. Closest to anticipation on the pass-code limbs: US 5,414,781 (reference 8) and US 5,245,676 (reference 3), both Xerox, both § 102(b) art, both disclosing pass-code-location extraction from CCITT-compressed data. They most directly implicate claims 2–4, 21–24, 28–30 and 37–38 (the pass-code / CCITT Group III–IV limitations), but they cannot anticipate those claims because the parents are not anticipated.
3. Closest to anticipation on the matching-metric limbs: the Huttenlocher et al. NPL (with Rucklidge) supplies the Hausdorff/modified-Hausdorff distance. It is not a "patent citation," but it is the single most on-point reference for the metric language in claims 1, 11, 21, 28, 32 and 36.
4. The real challenge is § 103, not § 102: the defensible combination is US 5,414,781 / US 5,245,676 / Spitz (pass codes as anchor locations in CCITT G4) in view of Huttenlocher et al. / Rucklidge (Hausdorff comparison and smallest-distance selection), optionally with US 5,465,353 (image matching against a document database) for the two-stage framework.

6. Caveats

• I located the specific patent 6,104,834; no results for other "834" numbers (e.g., 6,641,834; the Magnolia "'483" patent; the Centripetal '370/'917 patents) are included.
• Abstracts/full element-by-element reads were confirmed for references 3, 7, 8, 10 and for the four NPL documents; abstracts were not retrieved for references 4 (US 5,388,167) and 5 (US 5,586,196), and the remaining references were characterized from their titles and issuing-art context. Those assessments are flagged as provisional.
• § 102 categories are assigned on the printed filing/issue dates and the 1996-08-01 filing date; the Unified Patents ±1-day date variants do not alter them. For § 102(e) references, the effective prior-art date is the reference's own U.S. filing date, and pre-AIA § 103(c) common-ownership considerations could bear on the § 102(e) items (notably US 5,465,353, a Ricoh/Hull reference), which a full validity opinion should address.

I'll ground the key prior-art references before mapping them to the claims.

Obviousness Analysis — U.S. Patent No. 6,104,834 under 35 U.S.C. § 103

Patent: US 6,104,834 B1, "Matching CCITT compressed document images" (Inventor: Jonathan J. Hull; Assignee: Ricoh Company, Ltd./Ricoh Corporation; App. 08/695,825 filed 1996‑08‑01; granted 2000‑08‑15; expired 2016‑08‑01).
Sources for this section: the "Patent Citations (13)," "Non‑Patent Citations (8)," and "Family Cites" entries as they appear on the patent page, plus the authoritative claim text.

Contradiction flag (carried forward): my earlier summary recorded the patent's own NPL entry as "A. Lawrence Spitz, 'Skew Determination in CCITT Group 4 Compressed Document Images,' pp. 11 to pp. 25 (last reference 1989)." Independent literature (UNLV thesis, Univ. of Waikato thesis, DTIC report — links below) uniformly dates this Spitz paper to a 1992 Symposium on Document Analysis and Information Retrieval (Las Vegas), pp. 11–25. I report both literally. It is §102(b) art either way, so the discrepancy is immaterial to the analysis.

1. Legal framework and level of ordinary skill

Framework. Obviousness is assessed under Graham v. John Deere, 383 U.S. 1 (1966), and KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007): scope/content of the prior art, differences between the prior art and the claims, PHOSITA level, and objective indicia. Where the asserted combination is "of familiar elements according to known methods [yielding] no more than predictable results," it is obvious. KSR, 550 U.S. at 416. The specification's own admission of what was known (here, the CCITT two‑dimensional coding scheme and "pass codes," attributed to Hunter & Robinson, "International Digital Facsimile Coding Standards," Proc. IEEE (July 1980)) is usable as applicant‑admitted prior art for the motivation/reasoning enquiry.

PHOSITA. A person with a bachelor's or master's degree in electrical engineering or computer science and roughly 2–3 years of experience in document image analysis, facsimile/bilevel image coding, or image‑based information retrieval — or equivalent practical experience. Such a person would be familiar with the CCITT T.4/T.6 (Group III/Group IV) standards, run‑length/changing‑element coding, and standard pattern‑matching distance metrics.

Observation that frames everything below. The claimed advance is a combination: (a) a known class of features (CCITT pass codes / changing elements) known to fall at/near character bottoms, extracted directly from the compressed bitstream; and (b) a known comparison metric (the Hausdorff distance, including its modified/ranked variant), applied to select a nearest neighbor. Neither half is new; the novelty resides in the pairing. That is the classic §103 posture.

2. Mapping the independent claims to the cited art

References of low relevance (listed in the patent but bearing little on the claims): US 4,941,193 and US 5,065,447 (Iterated Systems — fractal/IFS compression); US 5,263,136 (Optigraphics — tiled multi‑resolution storage); US 5,583,144 (Xerox — anti‑counterfeit pattern detector); US 5,574,840 (Microsoft — text selection); US 5,586,196 (Sussman — document magnifier); US 5,623,679 (Waverley — notes); US 5,708,825 (Iconovex — automatic summarization/hyperlinks); US 5,388,167 (Hitachi — document image processing, general context). These are not needed for the combinations below.

3. Primary combination — renders claims 1–8, 11–18, 21–30, 32–38 obvious

Combination C1 = US 5,245,676 (Xerox/Spitz) + Huttenlocher et al. 1993 + Rucklidge 1994 (optionally + Hull 1995 / US 5,465,353).

What each teaches.

• US 5,245,676 discloses a computer‑implemented image analysis method that operates on CCITT Group 4 compressed data, locates the pass codes, and uses "the locations of the pass codes as the locations of said selected points of pixel color transition" (claims 3, 4, 12, 13). Its claim 5 defines the selected features exactly as this patent defines pass codes: color transitions on one scan line having "no corresponding pixel color transition … in an adjacent one of said scan lines." The EP 0 434 415 counterpart states that pass codes identify points "always located on the marks themselves," that white passes sit "at the bottom of each connected component," and that "there may be more than one fiducial point per mark" — i.e., a set of characteristic character anchor points. In short, C1 supplies the entire feature‑extraction half of every independent claim, including the compressed‑domain and pass‑code limitations, and does so in a single pass over the compressed bitstream.
  • https://patents.google.com/patent/US5245676
  • https://patentimages.storage.googleapis.com/ff/6b/d4/9b03a35528efcb/EP0434415A3.pdf
  • https://www.freepatentsonline.com/[5245676](/patent/5245676).html
• Huttenlocher, Klanderman & Rucklidge (1993) teaches comparing images/models using the Hausdorff distance, including the modified (k‑th ranked) Hausdorff measure that the '834 specification itself adopts verbatim for outlier robustness, and its use with binary images. It supplies the comparing step of claims 1, 11, 21, 28, 32, 36.
• Rucklidge (1994) is, by its title alone, directed to "Efficient Computation of the Minimum Hausdorff Distance for Visual Recognition" — supplying both (i) the "smallest Hausdorff distance" selection limitation found in every independent claim, and (ii) the efficiency/speed that the '834 specification repeatedly asserts as the advantage.
• Hull 1995 (the inventor's own cited paper) and US 5,465,353 supply the document‑retrieval architecture: extract a descriptor set per document, store descriptors for a corpus, match an input against the corpus, and return the best candidate (the '353 patent outputs "the document tags which have the most accumulated counts").

Motivation to combine — articulated per KSR.

1. Same field and same problem. Both US 5,245,676 and the '834 claims concern analyzing bilevel document images in the CCITT compressed domain; both exploit the same physical insight (pass codes cluster at character bottoms/baselines). A PHOSITA looking for a feature set to feed a matching system has an express reason to use the pass‑code feature set that the Xerox patent already established as a compact, character‑anchored, decompression‑free descriptor.
2. The problem Huttenlocher solves is the precise problem the '834 spec identifies. The '834 specification criticizes the original Hausdorff metric's outlier sensitivity and adopts the Huttenlocher modified metric to fix it. That admission is itself a teaching‑away argument inverted: the claimed use of Hausdorff is Huttenlocher, and the specification offers no improvement over it.
3. Predictable results / known technique for a known purpose. Substituting one well‑known distance metric for another in a nearest‑neighbor document‑retrieval loop (the loop already disclosed by Hull/US 5,465,353) is the paradigm of a predictable, design‑choice combination. KSR, 550 U.S. at 416–17, 421.
4. Reasonable expectation of success. The features are 2‑D point sets; the metric is defined on point sets; Huttenlocher demonstrates the metric on binary images; Rucklidge supplies a working efficient implementation. Nothing in the combination required the PHOSITA to discover a new principle.
5. The "unexpected" numerical result is already in the art. The specification's statement that "approximately 80% to 90% of the components have an attached pass code" is a restatement of what US 5,245,676 already explains (a white pass is guaranteed at the bottom of each connected component, and additional passes arise from aliasing on horizontal structures) and what Hunter & Robinson's standard explains mechanically.

Claim‑specific notes for C1.

• Claim 1 needs only: changing‑element detection (US 5,245,676 claim 5), capture of locations as anchor points (US 5,245,676 claims 3–4, 12–13), Hausdorff comparison against stored sets (Huttenlocher; stored sets per Hull/US 5,465,353), and smallest‑distance selection (Rucklidge). Every element is accounted for.
• Claim 36 is even easier: US 5,245,676's (x, y) coordinates of pass codes are a "two‑dimensional (2‑D) representation of extracted features."
• Claim 21 is a system claim whose only added structure is "a storage device for storing electronic representations of images of documents," disclosed by US 5,465,353 (descriptor database) and/or US 5,267,047 (facsimile subsystem in an image archiving system).
• Claim 28/32 (compressed input with pass codes) are squarely met by US 5,245,676's compressed‑data embodiment plus the Hausdorff comparison.

Verdict on C1: a strong prima facie case of obviousness across all six independent claims. The only limitation requiring a second reference beyond US 5,245,676 is the Hausdorff comparison/selection, and that is furnished directly by two references the applicant itself cited.

4. Secondary/alternative combinations

C2 — US 5,414,781 (Xerox, "document classifying") + Huttenlocher + Rucklidge. US 5,414,781 decodes Group 4 codes by character‑string recognition, accumulates horizontal displacement to obtain x, tracks a color state bit to identify white vs. black passes, and increments a line counter y at line end. This reference is unusually well matched to the dependent claim cluster:

• claim 5 family (horizontal accumulation of codeword pixels) — expressly taught;
• claim 6 family (vertical location) — the reference increments the line count at line end and can thereby index lines, and the patent's own claim 6 alternative (counting EOL indicators) is supplied by the CCITT Group III standard (Hunter & Robinson);
• claim 7 family (single pass) — the reference is explicitly a single sequential pass;
• claim 8 family (detecting changing elements) — the reference's "horizontal code"/"vertical code"/"pass code" classification is the changing‑element machinery.
  Used with Huttenlocher/Rucklidge, C2 makes the dependent claims obvious with essentially no gap.

C3 — Spitz 1992 (NPL) + Huttenlocher 1993 + Rucklidge 1994. The Spitz paper — reported in the patent's own IDS at pp. 11–25 — is the literature counterpart of US 5,245,676 and was the first proposal of a skew estimator for CCITT Group 4 compressed images (per UNLV's survey: "Spitz [21] was the first to propose such a skew estimator for CCITT Group 4 compressed images," https://oasis.library.unlv.edu/cgi/viewcontent.cgi?article=4230&context=rtds; dating corroborated at https://researchcommons.waikato.ac.nz/bitstream/handle/[10289/14909](https://assignmentcenter.uspto.gov/search/patent/reelFrameDetail?reelFrame=10289-14909)/thesis.pdf and https://apps.dtic.mil/sti/tr/pdf/ADA458239.pdf). It supplies the compressed‑domain pass‑code feature extraction with the same motivation (speed, no decompression).

C4 — Baird 1987 + Huttenlocher + Rucklidge + Hull 1995. For claims 9/10/19/20/31/35 ("anchor points … indicative of approximate locations of characters" / "anchors of characters"), Baird's fiducial points at the bottom‑center of each character bounding box — as described in the US 5,245,676 background — are character anchors in the literal claim sense. This supplies the character‑anchor limitation independently of the pass‑code route.

C5 — US 5,465,353 (Ricoh/Hull) + Huttenlocher + Rucklidge. For the "compare to sets previously obtained for prospective matching documents" and "identify best match" elements, US 5,465,353 provides the multi‑reference descriptor database and best‑candidate output; Hull 1995 provides the document‑image‑specific distortion‑tolerance rationale. Notably, an EPO search report on a related Ricoh/Hull family member (EP 1 555 626) cited US 5,465,353 and Hull 1995 as "X" (alone‑anticipatory) documents against document‑retrieval claims of the same genus — corroborating that the retrieval architecture here was not novel (https://patentimages.storage.googleapis.com/3c/34/49/24363b96768550/EP1555626A3.pdf).

Context references for the "facsimile receiver" embodiment (specification: the invention "may be easily embodied in a facsimile receiver where the appropriate compressed representation is already available"): US 5,267,047 (IBM — facsimile subsystem in an image archiving system) supplies the motivation/expectation of success for doing document operations on already‑compressed fax data inside an archiving appliance.

5. Dependent-claim disposition (summary)

No dependent claim introduces an element that is not accounted for by the cited art + the PHOSITA's ordinary knowledge.

6. Where the obviousness case is weakest (be candid)

1. "Smallest Hausdorff distance" as an express selection step. Huttenlocher's 1993 paper is framed largely as a binary decision test (h_k(A,B) ≤ δ). The express notion of ranking candidates and selecting the minimum comes from Rucklidge's title/abstract (minimum Hausdorff distance) and from the general nearest‑neighbor retrieval paradigm (US 5,465,353's "most accumulated counts"). The combination is still strong, but if Rucklidge were excluded, one would need to argue that selecting the minimum of a distance metric is an obvious design choice — which is straightforward under KSR but is an argument, not an express disclosure.
2. US 5,245,676's stated purpose is skew estimation, not matching. A challenger must articulate the motivation to repurpose the pass‑code feature set from skew measurement to document identification. This is a KSR "different application of a known technique" argument and is well supported by (i) the overlapping problem of robustly locating character features, (ii) Hull 1995/US 5,465,353 already doing document matching on extracted features, and (iii) the reference's own statement that the pass codes lie on the marks themselves.
3. Claim 1's breadth cuts both ways. Claim 1 does not require pass codes or CCITT at all — only "changes in pixel values between successive pixels." That breadth makes it easier to invalidate (Baird fiducial points / any changing‑element extraction + Hausdorff) but also means the claim's validity may turn on whether Hausdorff was a non‑obvious choice for this feature type. Given that the applicant cited Huttenlocher in its own IDS and adopted its modified metric wholesale, that argument is weak.
4. Objective indicia are absent. The record shows no litigation and no third‑party challenges (Unified Patents page for the patent lists no litigation and no opposition documents), the patent expired in 2016, and no evidence of nexus‑bearing commercial success, licensing, or copying appears in the cited materials. There is therefore no secondary‑considerations evidence to rebut the prima facie case. (Caveat: absence of litigation is not evidence of validity either way; it simply means there is nothing for a patentee to point to here.)
5. Applicant‑admitted art must be used carefully. Hunter & Robinson (1980) is cited in the specification, not in the IDS list, and reliance on it is as applicant‑admitted prior art (AAPA) — permissible for the reasoning/motivation enquiry but best used in combination with the issued patent references (US 5,245,676, US 5,414,781) rather than as standalone ‑102 art.

7. Literalism / identification caveats (per operating rules)

• Claim 12 reads "The product of claim 7." Claim 7 is a method claim, so a product claim depending on it is internally inconsistent. If litigated, it would likely be construed as depending from claim 11. I do not silently correct it; the analysis above treats claim 12's substantive limitation (pass codes from a compressed representation) as derived from claim 11's product claim.
• Claim 27 recites "said pass word locations." The term appears nowhere else; literal reading creates an antecedent‑basis/112 defect. The corresponding limitations elsewhere recite "pass code locations." I report it literally and treat the intended subject matter as pass‑code locations for §103 purposes.
• US 5,245,676 vs. Spitz 1992. These are two distinct citations on the face of the '834 patent — an issued Xerox patent (Spitz named inventor on the EP counterpart) and a Spitz conference paper the patent lists at pp. 11–25 with a "1989" parenthetical that external sources render as 1992. Both are treated as prior art; I do not merge or re‑date them.
• Family members are not prior art. US 6,546,136 B1 (the continuation of this application) and JP 3766186 B2 appear in the "Cited By"/family listings but share the 1996‑08‑01 priority and cannot be §103 art against it. Likewise JPH09270902A (Ricoh; 1997‑10‑14 publication) postdates the 1996‑08‑01 critical date and is not §102(a)/(b) art.

8. Bottom line

Under § 103 and KSR, claims 1, 11, 21, 28, 32, and 36 — and their dependents — would have been obvious over US 5,245,676 in view of Huttenlocher, Klanderman & Rucklidge (1993) and Rucklidge (1994), with Hull (1995) / US 5,465,353 supplying the document‑retrieval framework and US 5,414,781 (and, for claim 6, the CCITT Group III EOL convention admitted via Hunter & Robinson 1980) supplying the horizontal‑accumulation, vertical‑location, single‑pass, and changing‑element dependent limitations.

The motivation is direct and articulated in the references themselves: both the '676 patent and the '834 claims mine the same CCITT pass‑code construct from the same compressed bitstream for the same reason (speed, memory efficiency, no decompression), and the Hausdorff metric the claims require is the very metric (original and modified) that the patent's own cited NPL literature contributed to the field. The strongest, genuinely arguable point of departure is the express "smallest Hausdorff distance" selection step, which is most cleanly supplied by Rucklidge (1994) and by the nearest‑neighbor retrieval teachings of US 5,465,353/Hull 1995.

Confidence: High that C1 establishes a prima facie obviousness case for the independent claims and the C2‑supported dependents. Moderate‑high on the "smallest distance" limitation absent Rucklidge. No objective indicia were located that would rebut the case.