US 5821945

I'll search for authoritative information on this patent, including any litigation or CAFC docket activity.

US Patent 5,821,945 — Summary

Caveat up front: The search did not surface any U.S. court docket, PTAB proceeding, or CAFC appeal (2026 or otherwise) involving patent 5,821,945. The only hits referencing the number were third-party references (e.g., an EPO search report for EP 01 30 2138 citing it as category "A" prior art, and IDS listings in unrelated patents). I could not directly query USPTO PatentCenter or the CAFC docket system in this session, so I cannot affirmatively rule out litigation — I simply found no authoritative record of any.

Bibliographic data (per Google Patents, patent/US5821945A/en, fetched 2026-09-27)

Note the two-key date situation: 1995-02-03 is the priority date; the 1997-05-15 date is the filing date of this continuation application. The specification explicitly states: "This is a continuation of application Ser. No. 08/382,877, filed Feb. 3, 1995 now U.S. Pat. No. 5,708,767."

Abstract (verbatim)

"A new technique for extracting a hierarchical decomposition of a complex video selection for browsing purposes, combines visual and temporal information to capture the important relations within a scene and between scenes in a video, thus allowing the analysis of the underlying story structure with no a priori knowledge of the content. A general model of hierarchical scene transition graph is applied to an implementation for browsing. Video shots are first identified and a collection of key frames is used to represent each video segment. These collections are then classified according to gross visual information. A platform is built on which the video is presented as directed graphs to the user, with each category of video shots represented by a node and each edge denoting a temporal relationship between categories. The analysis and processing of video is carried out directly on the compressed videos. Preliminary tests show that the narrative structure of a video selection can be effectively captured using this technique."

Independent claims — plain language

The patent has 16 claims, with two independent claims (claim 1, apparatus; claim 9, method). The remaining claims depend from them.

Claim 1 (video browser — apparatus): Covers a browsing tool with three cooperating elements:

1. Node-defining means that creates multiple nodes, where each node represents a collection of video shots (i.e., a group of shots treated as one scene-level unit).
2. Temporal-relationship-defining means that establishes at least one temporal relationship between at least a first and a second node (i.e., an ordering/sequence link between two scene groups).
3. A display that shows graphical representations of both the nodes and the temporal relationship(s).

Written in "means for" format, so it would be construed under 35 U.S.C. § 112(f) as covering the corresponding structure in the specification and equivalents.

Claim 9 (method for presenting video shots for browsing): The method counterpart of claim 1 — the steps of (a) defining a plurality of nodes each representing a collection of video shots, (b) defining at least one temporal relationship between at least first and second nodes, and (c) displaying graphical representations of the nodes and the temporal relationship.

Key dependent-claim concepts:

• Claims 2/10 — shots within a node interrelated by color, spatial correlation, and/or shape (the visual-similarity clustering basis).
• Claims 3/11 — shots within each node arranged in temporal order.
• Claims 4/12 — the temporal relationship displayed as an edge.
• Claims 5/13 — displayed as a hierarchical scene transition graph.
• Claims 6/14 — the hierarchy has multiple levels, each constructible according to different criteria (the "act/scene/shot" multi-level organizing idea).
• Claims 7/15 — means/steps to group shots of two or more nodes into an additional node.
• Claims 8/16 — means/steps to ungroup shots from a node to form an additional node.

The user-interaction grouping/ungrouping claims (7–8, 15–16) map to the specification's interface for re-arranging nodes and regrouping clusters.

Technical context (from the specification)

The disclosed pipeline operates on DC sequences extracted directly from Motion JPEG and MPEG compressed video (no full decompression), comprising four steps: shot identification (using the Yeo & Liu rapid scene-analysis algorithm, incorporated by reference), clustering of visually similar shots, presentation of content/structure via a scene transition graph, and hierarchical organization of graph levels. Similarity is measured via color histogram intersection (Swain & Ballard), moment invariants for shape, and luminance-projection correlation. Clustering is hierarchical (Jain & Dubes proximity-matrix framework), with under-classification preferred over over-classification.

Uncertainty notes

• No litigation, PTAB, or CAFC 2026 activity for this patent was found in the searches performed. This should not be read as confirmation that none exists — a definitive answer requires a direct USPTO PatentCenter and CAFC docket query.
• The claim language quoted from the source rendering contains a couple of apparent OCR/figure artifacts (e.g., "collecting of video shots" in claim 1 reads as "collection" elsewhere, and "claim. 1" in claim 4). I have preserved the source text but flag these as likely transcription errors rather than deliberate claim wording.

I'll search for litigation involving US patent 5821945.

Let me search litigation-specific databases more specifically.

Let me do a couple more targeted checks on litigation databases.

Findings: Litigation involving U.S. Patent No. 5,821,945

Result: No litigation involving U.S. Patent No. 5,821,945 was found.

I searched litigation-oriented sources (Unified Patents litigation portal, RPX Insight, CourtListener, Justia Dockets, and general web searches keyed specifically to the exact number "5821945" / "5,821,945"). Every hit for that exact identifier resolved to one of the following non-litigation contexts:

• The patent document itself — US5821945A, "Method and apparatus for video browsing based on content and structure," inventors Boon-Lock Yeo, Minerva M. Yeung, Wayne Wolf, and Bede Liu; assignee Princeton University; filed 1997-05-15; granted 1998-10-13; priority 1995-02-03; status Expired – Fee Related (anticipated expiration 2015-02-03).
• Prior-art citations — the patent is cited as a reference in later patent prosecution (e.g., EP1132835A1, EP1722371A1, EP1067786B1, EP1081960B1, WO2008065297A1, US10757481). Being cited as prior art is not litigation.
• PTAB/patentability search reports — the number appears in USPTO PTAB petition search-term tables (petitions 1556542/1556543), i.e., as a searched patent number, not as a case.
• Same-number confusions (excluded per your instruction) — these are different identifiers and were not treated as hits:
  • U.S. Patent No. 6,558,194 ("194 Patent") — PPC Broadband v. Corning Optical Communications (N.D.N.Y./E.D. Wis.) — different patent.
  • JP5821945B2 — a Japanese patent for an "Upper limb training device."
  • Ukrainian court decision No. 5821945, and a Lithuanian registry notice No. 5821945 — unrelated civil/administrative matters.
  • SRAM v. Princeton Carbon Works — the assignee "Princeton" here is a bicycle-wheel maker, not Princeton University, and the asserted patents were US10611188 / US9610800.

Case table

Notes and caveats

1. No district court, Court of Appeals for the Federal Circuit, ITC, or PTAB proceeding naming U.S. 5,821,945 as an asserted or challenged patent could be located. It also does not appear as an asserted patent in the litigation records indexed by Unified Patents or RPX for the relevant period.
2. Related family: U.S. 5,821,945 is a continuation of application Ser. No. 08/382,877, filed 1995-02-03, which issued as U.S. Pat. No. 5,708,767 (also Princeton University). If you want a complete enforcement picture for this family, the '767 patent is the companion to check — but I did not find litigation on 5,821,945 itself.
3. Confidence level: High that no frequently-indexed litigation exists; moderate overall, because litigation can exist in dockets not captured by free portals (e.g., pre-2000 cases, sealed matters, or non-indexed state proceedings). A definitive negative would require a paid docket search (Docker Navigator / Lex Machina) run on the exact patent number and its entire family, plus a review of Princeton's patent-assignment/licensing records.
4. Consistent with the "operating rules," I did not auto-correct or conflate the number: any hit for "6,558,194," "JP5821945," or "5821945" as a non-patent identifier was treated as a different identifier and excluded.

If you'd like, I can next run the same check against U.S. Pat. No. 5,708,767 and any other members of this Princeton family, or attempt a PACER/CourtListener docket search on the exact string.

PTAB Proceedings — US 5,821,945 ("Method and apparatus for video browsing based on content and structure")

Patent snapshot (from the authoritative Google Patents text supplied): US 5,821,945 A; application US 08/856,899, filed 1997-05-15 as a continuation of Ser. No. 08/382,877 (now US 5,708,767), priority 1995-02-03; granted 1998-10-13; original and current assignee The Trustees of Princeton University (Google Patents lists "Princeton University"); 16 claims (claims 1–8 apparatus / "video browser," claims 9–16 method); legal status Expired – Fee Related, with an anticipated expiration date of 2015-02-03. Sole examiner-cited references: US 5,231,487 (Sony Broadcast), US 5,329,596 (Hitachi), US 5,521,841 (Siemens), and the parent US 5,708,767 (Princeton).

Proceedings overview

Total AIA trial proceedings on file: 0 — no IPRs, no PGRs, no CBMs, no derivation proceedings, and (per the structured USPTO ODP block supplied with this task) nothing indexed as active, settled, or denied. Breakdown: 0 active, 0 claims invalidated, 0 claims sustained, 0 settled, 0 institution denials. The defensive posture this gives a defendant is not "hardened by surviving IPRs" and not "claims canceled" — it is simpler and stronger: all 16 claims are UNTESTED at the PTAB, and the patent's term expired on 2015-02-03. There is no live patent right to invalidate, so the practical defense is not an IPR strategy but a standing/damages one.

I found no IPR/PGR/CBM petition naming this patent in web sources either. Searches for "5821945" plus IPR/CBM/PTAB terms returned only unrelated reexaminations of other patents (e.g., control nos. 95/000,679 and 95/002,245 for US 6,746,885), never a proceeding on the '945 patent. The Google Patents record's own "Citations / Cited By" tables (which flag PTAB-related prosecution events when they exist) show no post-grant proceeding entries for US 5,821,945: https://patents.google.com/patent/US5821945/en.

No proceedings to enumerate

There is no proceeding number, petitioner, panel, institution decision, Final Written Decision, settlement, or appeal to report for this patent. Per the operating rules, I am not inventing one. Everything in the "for each proceeding" template would be fabricated if I filled it in.

Two structural reasons the empty docket is unsurprising, and one caveat:

1. Statutory availability window was tiny. The '945 patent is pre-AIA (priority 1995-02-03), so PGR was never available and CBM was inapplicable on its face (video browsing/story-structure analysis is not a "financial product or service"). IPR became available 2012-09-16; the patent expired 2015-02-03. That is roughly a 29-month window in which an IPR could have been filed at all — and only against a patent with under three years of remaining life.
2. No assertion found. I searched for infringement litigation by Princeton on this patent and found none in the sources searched. Patents that are never asserted rarely attract IPRs; the absence of PTAB activity here is consistent with a patent that was a research artifact rather than a litigation asset. (Caveat: I did not run a full PACER/Docket Alarm litigation sweep across all 94 districts and all years, so treat "never asserted" as "no evidence found," not as a verified negative.)
3. Caveat on completeness. My verification is the structured ODP block (authoritative: no proceedings) plus web searches. I cannot rule out a very recent filing that neither ODP nor search engines have indexed. If that matters to you, confirm in PTAB E2E / Patent Center directly before relying on the zero count.

Strategic summary

Claim status: 16 of 16 claims UNTESTED; 0 canceled; 0 sustained; 0 narrowed. Claims 1–8 (the "video browser" apparatus claims, including the user-grouping dependent claims 7–8) and claims 9–16 (the method counterparts, including dependent claims 13–14 on the hierarchical scene transition graph and 15–16 on grouping/ungrouping) stand exactly as they issued on 1998-10-13. There is no IPR certificate, no certificate of correction narrowing them, and no reexamination certificate that I could find. If someone waves a claim chart at you built on claim 1 or claim 9, the chart is not backed by any adjudicated validity — but it is also not pre-empted by any PTAB estoppel, because no estoppel exists.

Estoppel landscape: none, in either direction. 35 U.S.C. § 315(e)(2) estoppel attaches only to a petitioner that obtained an institution decision; there is no petitioner here, so no party is barred from anything. Symmetrically, the patent owner has forfeited nothing — but it also has nothing left to enforce, which is the far more important point. Because the term ran out on 2015-02-03, and 35 U.S.C. § 286 caps damages recovery at six years before suit, the last date on which a damages claim on the '945 patent could have been filed and still reached pre-expiration infringement was roughly 2021-02-03 (subject to tolling/disability doctrines). Any demand letter or complaint served in 2026 is outside the damages window on its face.

Pattern signals: none to speak of. No serial petitioner, no defensive aggregator (Unified Patents, RPX, etc.) in the chain — those entities target asserted, in-force patents, and this one was neither. The patent owner (Princeton) has not pursued PTAB appeals on this patent because there is nothing to appeal. The only family-relevant wrinkle is the parent, US 5,708,767 (issued 1998-01-13), which shares the specification and priority chain; if you are evaluating a portfolio theory-of-harm, the '767 is a separate patent with its own (also likely expired) term and its own docket, and should be checked separately rather than assumed to share this one's status.

Bottom line for a defendant: the correct response to an assertion of US 5,821,945 in 2026 is not a validity defense — it is 35 U.S.C. § 271(a) timing plus § 286 limitations plus "the patent expired 2015-02-03 and no PTAB proceeding ever disturbed or needed to disturb it." There is no FWD to quote at you and no canceled claim to point to.

Recommended next steps

• Do not budget for an IPR. There is no institution decision, no FWD, and no § 315(e)(2) estoppel to leverage, so an IPR-based defense has no purchase here. Confirm the empty docket yourself at USPTO PTAB E2E (https://ptacts.uspto.gov/) by searching the patent number, and confirm term status in Patent Center (https://patentcenter.uspto.gov/) — the Google Patents record shows legal status "Expired – Fee Related" with anticipated expiration 2015-02-03, and you want the maintenance-fee / expiration event history from the Office itself: https://patents.google.com/patent/US5821945/en.
• Build the § 286 / expiry argument instead. Pin down the exact expiration and any lapse event, then compute whether any asserted act falls within the six-year lookback measured back from the filing date of whatever pleading you have received. On the dates here, the arithmetic is unfavorable to the patent owner.
• Check the parent and siblings before you respond. US 5,708,767 (the Ser. No. 08/382,877 parent) and any other continuations in the family each have independent terms and dockets. Run each through PTAB E2E and Patent Center; do not assume the '945 result carries over.
• Sweep for past assertions even though the patent is expired. If the patent was asserted against anyone between 2009 and 2021, those records (PACER, CourtListener RECAP at https://www.courtlistener.com/) will tell you whether licensees, covenants-not-to-sue, or prior invalidity contentions exist — useful for a knowledge/notice or collateral-estoppel narrative even though the patent itself is dead.
• If a demand letter arrives citing claims 1–8 or 9–16: note in your response that no claim of US 5,821,945 has ever been invalidated or affirmed by the PTAB because no AIA trial was ever instituted, and that the patent expired 2015-02-03 — so the letter cannot be supported by any live infringement theory.

If you have a proceeding number that is not reflected in the ODP block or in the searches above, give it to me and I will pull the petition, institution decision, and any FWD from PTAB E2E and update this analysis claim-by-claim.

I'll research the assignment record for US 5821945 before building the report.

Let me dig deeper into the assignment/legal-events record.

Let me check for litigation and any assignment records on the parent patent.

Research note and source limitation

I was able to reach the Google Patents legal-events mirror of the USPTO assignment record (https://patents.google.com/patent/US5821945A/en) and third-party mirrors (uspto.report/patent/grant/5821945, FreePatentsOnline, PatentLeaderboard), but I could not execute a live query against https://assignmentcenter.uspto.gov/ or https://assignment.uspto.gov/patent/index.html directly in this session. Take the assignment findings below as mirror-derived and verify the reel/frame entries manually at the Assignment Center search page before relying on them.

What the mirror shows for US 5821945: the only title event on record is the original inventor→university assignment reflected in the applicant entry "1997-05-15 — Application filed by Princeton University." There is no post-issuance assignment, no security agreement, no merger, and no change of name in the legal-events record. The events listed are only: filed by Princeton (1997-05-15), priority to 08/856,899 (1997-05-15), granted (1998-10-13), published (1998-10-13), anticipated expiration (2015-02-03), status Expired – Fee Related.

Inventors

Employer attribution is determinable from the record: the application was filed by and assigned to The Trustees of Princeton University, and the specification's own non-patent citations include "The Princeton video library of polities" (Digital Libraries '94) — the Princeton digital-library research project that produced this work. The inventors are named in the same order on the parent US 5,708,767 (application 08/382,877, filed 1995-02-03), which shares the identical title and inventive entity.

Unusual patterns: None of the classic red flags. There is no evidence of inventors departing the assignee within 12 months of filing, no inventor-side assignment to a third party (which did not exist as an issue here — the parent predates Stanford v. Roche, and title sat with Princeton), and no split inventorship. This is an academic research group; the natural later career moves (Yeo and Yeung to industrial research labs, Wolf to another university) are ordinary academic mobility and are not tied to any recorded assignment — I flag them only so they are not mistaken for a fire-sale pattern. I could not verify post-filing career moves from primary sources in this session, so treat that last sentence as background, not a finding.

Original assignee

The Trustees of Princeton University — the entity named on the issued patent (Google Patents lists "Princeton University" as both original and current assignee; FreePatentsOnline lists "The Trustees of Princeton University" as assignee of record).

• Line of business: nonprofit higher education and sponsored research. Not a commercial producer of video-browsing software.
• Product embodying the claims: No commercial product. The specification describes a research prototype — a hierarchical scene-transition-graph browser implemented and benchmarked on an SGI Indy against a 16-minute 1992 Democratic Convention clip and a news report. The specification itself (Conclusions) frames this as feasibility demonstration, not a shipped product.
• Current status: Operating. Princeton University is an ongoing institution; there is no dissolution, acquisition, or bankruptcy event anywhere in the record.
• Research provenance: The cited project work ("Princeton video library of polities," "The digital video library system," Digital Libraries '94) is consistent with the NSF/ARPA/NASA Digital Libraries Initiative era. That implies possible U.S.-government funding and Bayh-Dole obligations on Princeton's side, but the patent text does not state a funding source and I could not verify one — do not treat this as established.

Assignment timeline

No recorded assignments. The USPTO assignment record, as mirrored in Google Patents' legal events, contains no post-issuance assignment entries for US 5821945 — no reel/frame, no conveyance type, no assignor/assignee pair beyond the original inventor→Princeton filing, and therefore no correspondent of record to capture. The only name that appears as owner, from application filing (1997-05-15) through expiry, is Princeton.

Per the operating rule for this situation: this is itself the finding. There is no chain to reconstruct. The inventor→Princeton assignment executed around the 1995-02-03 priority filing is the sole title event, and no reel/frame could be retrieved to cite. Verify manually at the Assignment Center — if a reel/frame does exist for that initial assignment, it will be the only entry, and it will point to The Trustees of Princeton University.

Timeline diagram

timeline
    title Ownership of US 5821945
    1995 : Priority application filed by Princeton
    1997 : Continuation application filed
    1998 : Patent issued to Princeton
    2015 : Patent expired

I deliberately did not add a "no assignment recorded" node, because inventing an event label would misrepresent the record; the diagram reflects only events that appear in the legal-events data.

NPE / troll-pattern signals

1. Shell-entity transfer — not present. There is no transfer of any kind. Assignee of record has been The Trustees of Princeton University throughout (Google Patents applicant entry, 1997-05-15; status Expired – Fee Related, 2015-02-03). No "IP / Holdings / Ventures" entity appears anywhere in the chain.

2. Known asserter in the chain — not present. No entity matching the named NPE list (Acacia, Marathon, IV, IPNav, Wi-LAN/Mosaid-Conversant, Vringo, Pendrell, Innovatio, MPHJ, Lumen View, Round Rock, Spangenberg entities, etc.) appears as assignee. The only assignee is a university. I found no RPX or Unified Patents directory entry linking this patent number to an assertion campaign.

3. Repeat correspondent across the chain — not applicable / not present. With zero recorded transfers, there is no correspondent to trace. This signal cannot fire on a single-assignment chain, and I decline to name a correspondent I could not read from a reel/frame.

4. Cascading transfers — not present. No transfers at all, so no chained-LLC sequence and no shared-correspondent address pattern. The patent was cited as prior art by 181+ later patents (Sharp, Microsoft, Sony, Canon, Columbia, ActiveVideo, Nielsen, Disney, etc.) — that is a citation footprint, not an ownership chain, and citations do not indicate NPE activity.

5. Pre-litigation transfer — not present. I found no infringement suit naming US 5821945 in the sources I could reach, and no assignment dated within 6 months of any suit. Note the practical point: with a 1998-10-13 grant and a record listing anticipated expiration of 2015-02-03, the assertion window closed roughly a decade ago.

6. Bankruptcy fire-sale — not present. No Chapter 7/11 proceeding involving Princeton, and no sale-of-portfolio event. The assignee is a solvent university.

7. Privateering — not present. No operating company → NPE transfer exists to characterize as privateering, and no SEC filing or Patent Progress/EFF coverage surfaced.

8. Defensive aggregator — not present. The chain does not terminate at RPX, AST, LOT, Unified, or OIN. The inverse outcome occurred, but by a different mechanism: the patent simply lapsed (status "Expired – Fee Related," anticipated expiration 2015-02-03). A lapsed university patent is effectively dedicated to the public, but that is term/fee lapse, not a defensive-aggregator acquisition — do not record it as one.

Verdict

Insufficient data — in the sense contemplated by the taxonomy, i.e. no records beyond the original assignment.

Justification: the assignment record shows a single title event (inventor → The Trustees of Princeton University, reflected in the 1997-05-15 "Application filed by Princeton University" entry) and no post-issuance assignment whatsoever; every one of the eight NPE signals is therefore unsupported rather than merely unproven. Princeton is an operating nonprofit that never shipped a commercial product embodying the claims, but it also never asserted the patent commercially, and the patent carrier ended with the 2015-02-03 anticipated expiration and "Expired – Fee Related" status. There is no basis to call this an NPE chain; absent a manual reel/frame pull at the Assignment Center, the honest verdict is insufficient data.

Verify at: https://assignmentcenter.uspto.gov/ (search by patent number 5821945) and cross-check the legal-events pane at https://patents.google.com/patent/US5821945A/en and https://uspto.report/patent/grant/5821945.

Caveats on this report: (a) I could not query the Assignment Center directly this session, so "no recorded assignments" is a mirror-based conclusion pending manual confirmation; (b) no reel/frame numbers or correspondent names are cited because none were retrievable — I did not invent any; (c) inventor career moves after filing are outside the patent record and are flagged as such.

I'll research this systematically, starting with the patent itself and each cited reference.

Let me verify the parent patent and the key non-patent references' dates.

Prior-Art Analysis — US Patent 5,821,945

0. Confirmation of the searched identifier

I searched specifically for the number 5821945. It resolves to exactly one U.S. grant:

No results for near-numbers (e.g., 5,821,494 / 5,821,954) are reported below.

Critical art-date rule for the § 102 analysis. Because this application is a continuation of Ser. No. 08/382,877 (filed 1995-02-03), the claims are entitled to the 1995-02-03 effective filing date, not the 1997-05-15 filing date. Any reference must therefore predate 1995-02-03 to be § 102 prior art. The 1997 and 1998 dates appearing on the face of granted citations (e.g., US 5,708,767's 1998-01-13 grant) are publication dates, not art dates — the underlying applications' filing dates govern under pre-AIA § 102(e).

1. Summary conclusion

No reference cited on the face of US 5,821,945 anticipates any of its claims under 35 U.S.C. § 102. The claim set is narrow and specific: independent claim 1 (apparatus) and claim 9 (method) each require (a) nodes each representing a collection of video shots, (b) a defined temporal relationship between at least two of those nodes, and (c) a display of graphical representations of the nodes and the temporal relationship. None of the four patent citations, and none of the twelve unique non-patent citations, discloses all three elements in a single reference. The cited art is, at most, § 103 combination material, and the most relevant art is a different Siemens reference (US 5,521,841) plus the Arman and Zhang non-patent papers.

I flag this explicitly because the citation list is short (4 patent references, of which one is the parent) and the examiner clearly used the non-patent literature as the substantive prior art — the NPL citations are where the real § 103 risk lies.

2. The four patent citations (all marked "cited by examiner")

2.1 US 5,231,487 A — "Storage of video signals"

Brief description. This is a video compression-and-storage patent, not a browsing patent. It discloses an apparatus for storing a video signal using a two-dimensional sub-band spatial-frequency filtering arrangement, a quantizer that applies different quantization to the sub-band carrying DC luminance information than to the remaining sub-bands, an entropy encoder, and a storage arrangement (including error-correction coding and repeated recording of the DC sub-band on magnetic media). Its claim 1 is directed to the compression-plus-storage combination; claims 9–14 add error-correction and multiple-recording variants.

Anticipation analysis. None of claims 1–16 is anticipated. The reference contains no teaching of: (i) partitioning a video into shots; (ii) nodes each representing a collection of shots; (iii) a temporal relationship (edge) between nodes; or (iv) a display of node/edge graphical representations. It concerns intra-image decorrelation and quantization — an entirely different problem space. Its only conceivable relevance is background (DC/compressed-domain video data), which is not a claimed element of the '945 claims. Cited, in my assessment, for general context or examiner search completeness rather than substantive relevance.

2.2 US 5,329,596 A — "Automatic clustering method"

Brief description. A pattern-recognition clustering patent. It discloses a learning-mode method that partitions sample data of known categories into classes and subclasses in an n-dimensional space, generates "standard patterns" (centroids) per class, iteratively adjusts the standard-pattern positions, and stores the resulting tree-structure representation in memory. In a recognition mode, unknown input data points are routed down the tree to the nearest standard pattern until the lowest-level subclass is reached, with distance thresholds used to reject non-matching input. FIG. 9 expressly works an example with facial images.

Anticipation analysis. This is the most conceptually adjacent patent citation, because the '945 specification adopts a hierarchical-clustering framework (Jain & Dubes proximity matrices; "hierarchical clustering over partitional clustering") and a multi-level graph hierarchy. But it does not anticipate:

• Claim 1 / claim 9 — no video shots, no "collection of video shots per node," no temporal relationship between nodes, no display of a node/edge graph. US 5,329,596 builds an abstract classifier tree, not a temporal scene-transition graph.
• Claims 2 / 10 (shots interrelated by color, spatial correlation, shape) — the reference computes distances between generic n-dimensional data vectors; it does not disclose the three named low-level vision measures applied to video shots.
• Claims 5 / 13 (hierarchical scene transition graph) — the reference discloses a classification tree, but not a scene transition graph with edges induced by temporal precedence.
• Claims 7–8 / 15–16 (grouping shots of two nodes into an additional node; ungrouping shots from a node) — the reference's class/subclass division is the inverse operation (splitting), and it involves no user-facing regrouping of video shots.

Its realistic role is § 103 art for the "hierarchical clustering + tree of clusters" concept, combined with a browsing reference.

2.3 US 5,521,841 A — "Browsing contents of a given video sequence" ← closest patent prior art

Brief description. A content-based video browsing system that pre-processes video off-line to (a) detect scene changes in compressed video to form video shots, and (b) construct Rframes (representative frames) as abstractions for each shot. During browsing, the user scrolls through Rframes, selects one, and the system detects all Rframes with a degree of similarity to the selected Rframe and presents them to the user in a size/scale representative of the degree of similarity. The background section surveys Tonomura's variable-speed, sampling-flash, rush, and time-space browsers, and discusses histogram-distance scene-change detection.

Anticipation analysis. This is the strongest patent reference, but it still does not anticipate claims 1 or 9:

• It discloses shot detection and grouping of similar key frames (Rframes) — arguably touching the "clustering" idea underlying claims 2 / 10.
• It does not disclose "node defining means … each node representing a collection of video shots" in a graph structure; it presents a flat set of key frames scaled by similarity.
• It does not disclose "temporal relationship defining means … between at least first and second ones of the plurality of nodes," nor edges. Its presentation is by similarity, explicitly not by temporal ordering between groups.
• It does not disclose a hierarchical scene transition graph (claims 5 / 13), multi-level different-criteria hierarchy (6 / 14), or grouping/ungrouping of shots among nodes (7–8 / 15–16).

Realistic role: § 103 primary reference (it supplies the shot-detection + similar-shot-grouping + user-browsing elements) that would be combined with a temporal-graph teaching.

2.4 US 5,708,767 A — same title, same assignee ← NOT prior art

Brief description. Substantively identical disclosure to the '945 patent (same specification describing the Hierarchical Scene Transition Graph model, DC-sequence processing, color/shape/correlation similarity, and the Jain & Dubes proximity-matrix clustering). It issued from the parent application.

Anticipation analysis. US 5,708,767 cannot be § 102 prior art against US 5,821,945. It is not "by another" under § 102(a), (e), or (f) — it shares the same inventors, the same assignee, and the same priority date, and the '945 patent claims § 120 benefit from it. The examiner's listing of it among "Patent Citations" is a family/continuation cross-reference, not an art rejection. Any analysis that treated it as anticipating art would be legally erroneous. I flag this as the single most important trap in this citation list.

3. The non-patent citations (the substantive art)

The face of the patent lists 24 NPL entries, which collapse to 12 unique references (each appears twice in the Google rendering, once with and once without italics/markup). All predate 1995-02-03 except as noted.

Cross-check note. A separate Family/“Cited By” record for related Columbia University material lists an additional set of NPL items (e.g., Yeung et al., "Time-Constrained Clustering for Segmentation of Video into Story Units," ICPR 1996, and Yeung et al., "Video Browsing Using Clustering and Scene Transitions on Compressed Sequences," SPIE vol. 2417, 1995). Those are later than the 1995-02-03 priority date of the '945 patent and are therefore not prior art against it; they belong to the applicants' own subsequent publication stream.

4. § 102 claim-by-claim mapping (why nothing anticipates)

Bottom line: every cited reference fails at least one — usually several — of the three core claim-1/claim-9 elements. There is no single-reference § 102 case on this record.

5. Ranking of most relevant prior art

1. US 5,521,841 A (Siemens, 1994-03-31 / granted 1996-05-28) — closest patent art; supplies shot detection, representative-frame abstraction, similarity-based grouping, and a browsing interface. Best § 103 primary reference.
2. Arman, Depommier, Hsu & Chiu, "Content-based browsing of video sequences," ACM Multimedia 94 (Aug. 1994) — the printed-publication sibling of #1; same technical contribution.
3. US 5,329,596 A (Hitachi, 1991/1994) — hierarchical clustering with class/subclass tree and standard patterns; § 103 art for the multi-level hierarchy of claims 6/14.
4. Zhang, Kankanhalli & Smoliar, "Automatic partitioning of full-motion video," Multimedia Systems (Jul. 1993) — canonical shot-segmentation art.
5. Arman, Hsu & Chiu, "Image processing on compressed data for large video databases," ACM Multimedia (Aug. 1993) — compressed-domain processing.
6. Swain & Ballard (1991) and Hu (1962) — element-level art for the color/shape similarity recitations of claims 2/10.
7. US 5,231,487 A (Sony, 1991/1993) — peripheral; compression/storage only.
8. US 5,708,767 A — family member, not prior art.

6. Caveats and flags

1. Do not treat US 5,708,767 as prior art. It is the parent, same inventors/assignee, same priority date. Listing it in the "Patent Citations" table is a continuation cross-reference.
2. Effective date is 1995-02-03, not 1997-05-15. The 1997 filing date is the continuation's own filing date and does not move the art window. Conversely, the Yeo & Liu "Rapid Scene Analysis" paper, if assessed by its December 1995 issue date, postdates the priority date — but it is the applicants' own incorporated-by-reference work, and the face of the '945 patent lists it as "submitted … (unpublished)," so its status as § 102(b) art should not be assumed.
3. Minor date discrepancies across databases (flagged, not corrected): US 5,231,487 is listed with priority 1991-01-11 by Google Patents but 1991-01-10 (and grant 1993-07-26) by Unified Patents. Either way it is well outside the critical date.
4. The '945 patent is expired (fee-related, anticipated expiration 2015-02-03), so the practical stakes of this art analysis are historical/validity-record, not enforcement.
5. No litigation, PTAB, or CAFC activity for 5,821,945 surfaced in the prior section's searches; nothing in this search contradicts that, but it is still not a definitive negative.

Sources: https://patents.google.com/patent/US5821945A/en · https://patents.google.com/patent/US5521841A · https://patents.google.com/patent/US5329596A · https://patents.google.com/patent/US5231487 · https://patents.google.com/patent/US5708767A/en · https://uspto.report/patent/grant/5821945 · https://uspto.report/patent/grant/5521841 · https://www.freepatentsonline.com/5821945.html · https://portal.unifiedpatents.com/patents/patent/US-5231487-A

I'll research the prior art references listed in this patent before building the § 103 analysis.

Let me examine US5521841's full disclosure and the parent patent's claims, which are the two most probative references.

Obviousness Analysis — U.S. Patent No. 5,821,945 under 35 U.S.C. § 103

Building on the prior sections (bibliographic data, claim summary, litigation summary). I do not repeat that material; where I rely on it I reference it.

A. Framing and caveats

1. This patent is expired (anticipated expiration 2015‑02‑03, status "Expired – Fee Related" per Google Patents, https://patents.google.com/patent/[US5821945A](/patent/US5821945A)/en). The § 103 analysis below is therefore retrospective/hypothetical: its practical value is (a) understanding why the claims issued over the cited art in 1995–98, and (b) assessing the patent's defensive prior-art significance. No infringement action is possible on an expired patent, and no litigation was found (see prior Litigation summary).
2. Legal framework. At issuance (1998), pre‑AIA § 103 and the Graham v. John Deere / TSM methodology governed. If evaluated today, KSR Int'l v. Teleflex (2007) would apply, meaning the "motivation to combine" can be satisfied by "any need or problem known in the field" and by "design incentives" — a substantially lower bar. I flag both where they diverge.
3. Source rule. Per the operating rules, I treat the patent text provided as authoritative and interpret the identifiers literally. I did not conflate U.S. 5,821,945 with U.S. 6,558,194, JP5821945, or the non-patent "5821945" identifiers noted in the Litigation summary.

B. The Prior Art section: what actually qualifies

The "Prior Art" section of the page consists of 4 patent citations and 12 distinct non‑patent citations (the list repeats several entries; I de-duplicate). Not all of them are § 102 prior art against the '945 claims:

Key takeaway: The genuinely adverse references are US5521841A, US5329596A, Swain & Ballard, Jain & Dubes, Arman 1993/1994, Zhang 1993, and Szirmay‑Kalos. Everything else is either cumulative, the inventors' own work, or the family parent.

C. Level of ordinary skill in the art ("POSITA")

Given a Feb. 1995 priority date, a POSITA would have: a B.S. in electrical engineering or computer science (or equivalent) plus ~2–3 years of experience in digital image/video processing and multimedia systems, or an M.S. with focused coursework. That person would be familiar with (i) MPEG/JPEG DCT‑domain processing, (ii) color‑histogram and moment‑invariant image matching, (iii) standard clustering algorithms, and (iv) conventional graph‑drawing/visualization. All seven adverse references are squarely within that skill set — this supports the obviousness case and undercuts any "unexpected results" argument.

D. The independent claims (1 and 9) — element mapping and combinations

Claim 1 (apparatus) and claim 9 (method) recite the same three elements:

Combination 1 (primary rejection theory)

US5521841A ⊕ US5329596A ⊕ Arman 1994 ⊕ Szirmay‑Kalos 1994

This combination would support a prima facie case under pre‑AIA § 103, and a fortiori under KSR.

Combination 2 (alternative)

Arman 1994 ⊕ Zhang 1993 ⊕ Jain & Dubes ⊕ Szirmay‑Kalos

Here Arman is the base (Rframe browsing + shape/color icon management), Zhang supplies shot partitioning, Jain & Dubes supply hierarchical clustering to create the "nodes," and Szirmay‑Kalos supplies graph rendering. KSR makes this alternative at least as viable as Combination 1 because all references are in the same field of endeavor.

E. Dependent claims

F. Why a POSITA would have been motivated to combine (the KSR/TSM showing)

1. Same field, same problem, analogous art. All references address content‑based video browsing/indexing or image clustering. Every reference is from the video/multimedia/image‑processing arts and would have been considered by a POSITA working on the '945 problem.
2. Explicit, articulated need — the strongest motivation evidence. Arman 1994 states in terms: "none of the above systems address the problem of icon management. This is very important since as many as several thousand icons could be needed to represent the shots for each two hour video sequence." This is precisely the problem the '945 purports to solve ("a typical one hour program can produce hundreds of different shots … presenting shots in a one dimensional image array does not offer the users an effective and efficient means to browse"). When a reference itself frames the deficiency the patent claims to cure, the motivation to combine is essentially supplied by the art.
3. Predictable solution. Grouping similar items (clustering) and drawing a graph of their relationships (graph layout) were both mature, off‑the‑shelf technologies by 1995 (Jain & Dubes 1988; Szirmay‑Kalos 1994). There was a reasonable expectation of success.
4. Design incentive / finite number of options. Once a designer decides to present shot groups rather than a linear array, a graph (nodes + edges) is one of a small number of conventional presentation choices; KSR holds that "a court must ask whether the improvement is more than the predictable use of prior art elements according to their established functions."
5. Compressed‑domain processing (claimed implicitly in the spec at p.1, "All processing aspects … are performed on reduced data called DC‑sequence") is taught by Arman 1993 and Zhang 1993 — reducing any residual "efficiency" argument.
6. "No a priori knowledge" (the spec's stated advantage over Swanberg) is neutralized by US5329596 and Zhang 1993, which discover structure automatically without a priori models.

G. Where the § 103 case is weakest — the patentee's rebuttal

A rigorous analyst must flag that the prima facie case is not airtight:

1. No reference discloses the scene transition graph as such. The prior art's organizing axis is time (a one‑dimensional chronological Rframe/storyboard array — Arman, US5521841, Zhang) or similarity (Arman's "show me all similar Rframes"). The '945's asserted contribution is to re‑present the material in a directed graph whose edges encode temporal adjacency between clusters, so that the topology of the graph reveals narrative structure. None of US5521841 or Arman teaches edges drawn between shot groups; Szirmay‑Kalos teaches how to draw a graph but is agnostic about what the nodes and edges mean. This is a classic "teach the algorithm, not the application" gap that a patentee would exploit.
2. No teaching of graph topology as a source of semantics. The '945 claims (via claims 5–6/13–14) that the hierarchy and topology of the graph condense "repeatedly appearing shots" and reveal "act/scene/shot" structure. No cited reference uses graph topology for that purpose.
3. Claims 7–8 / 15–16 (user grouping/ungrouping) are thinly supported by the cited art. US5521841 discloses user‑driven similarity grouping, but the re‑forming of nodes by user action is not squarely taught. Under pre‑AIA TSM, a rejection here would be vulnerable.
4. Claim 6/14's "different criteria at each level" (visual cues below, semantics above) is a specific design choice that the cited art does not place in a single multi‑level graph.
5. The '945 relies heavily on claimed means‑plus‑function structure. Under § 112(f), "node defining means" and "temporal relationship defining means" are limited to the corresponding disclosed structure (the clustering engine and graph‑construction module, per FIG. 1 / FIGS. 9–12) and equivalents. This narrows the functional reach and is a defense the patentee could invoke to avoid prior‑art structures that group shots differently.

Net assessment: A rejection would likely succeed on claims 2/10, 3/11, 4/12, 5/13 and probably 6/14, and would be contestable on 1/9 (depending on how "temporal relationship … between nodes" and "display" are construed against Arman's temporal ordering + Szirmay‑Kalos's graph rendering), and weakest on 7/8 and 15/16. In other words: the heart of claim 1 — a display of high‑level content structure as a graph — is the most defensible concept, while the peripheral claims map cleanly onto the cited art.

H. Secondary considerations (objective indicia)

• No litigation, PTAB, or CAFC record was found involving the '945 (see prior Litigation summary). Consequently there is no objective evidence of nonobviousness (no nexus, no copying, no licensing, no commercial success) attributable to the challenged claims.
• Heavy forward citation — the page lists 181 "Cited by (examiner)" entries and a separate 311 "Cited By" list, including the Siemens follow‑on US6278446 and numerous Sharp Laboratories summarization patents. This is sometimes advanced as evidence of "recognition by others." It should be treated cautiously: citation is not a legally recognized secondary consideration, and much of the citation volume reflects the patent's role as a background reference in the video‑summarization field rather than praise for the claimed invention.
• No teaching away by the cited art: Arman's observation that "color histograms are a many‑to‑one mapping function" (and hence insufficient alone) does not teach away from the '945's combined color+correlation+shape similarity, which is exactly the multi‑feature approach Arman itself endorses.

I. Bottom line

The combination with the strongest motivation is US5521841A (Siemens) as primary reference, in view of Arman et al. 1994 and Szirmay‑Kalos 1994, because Arman expressly identifies the unresolved icon‑management problem that the '945 claims to solve — the textbook articulation of a "motivation to combine."

J. Explicit uncertainties

1. I could not retrieve the complete claim set of US5521841A in this session (the search results returned its abstract and portions of description/claims of a related Siemens application, US5835163). My characterization of US5521841 as disclosing similarity‑based grouping and temporal ordering of Rframes is supported by its abstract ("presenting the similar Rframes to the user in a size or scale representative of the degrees of similarity") and the Arman 1994 paper, but I have not verified that US5521841 expressly discloses edges between groups. If it does not, claim 1's § 103 case weakens correspondingly.
2. I did not exhaustively verify the US5329596A disclosure of video‑specific clustering; its specification (per Google Patents) uses facial images as an example, and I have assumed a POSITA would apply its tree‑clustering method to video shot features. That assumption is reasonable but is an inference, not a quotation.
3. US5708767A (the '767 parent) is treated as not prior art on the strength of the specification's own continuation statement. If the '945's claims were ever held not entitled to the '767's Feb. 3, 1995 benefit date, the '767 would drop into the prior‑art mix (as § 102(e) art) and would by itself disclose substantially the entire claimed subject matter — that contingency should be noted as the single largest validity risk.
4. I could not run the follow‑up searches (Szirmay‑Kalos full text; Zhang 1993 full text) before reaching the session's tool‑call limit; the characterizations above rest on the reference titles/venues and general knowledge, not on fetched full text.

Caveat per the operating rules: where these statements rest on general training knowledge rather than a fetched source, I have said so. Nothing here should be read as a legal opinion or as a substitute for a full validity search of the '945's complete family.