US 8311394

I'll search for authoritative information on this specific patent number.

Let me run a couple of targeted searches for litigation and docket activity.

US 8,311,394 B2 — Summary

Note on sources: The authoritative full text supplied from Google Patents (fetched 2026-09-28) is the primary basis for this summary. I ran three web searches (USPTO/Google Patents/FPO text, and targeted queries for "8311394" together with "CAFC 2026 docket," "litigation," and "appeal"). No litigation, PTAB proceeding, or Court of Appeals for the Federal Circuit docket for this patent surfaced in any search. I cannot confirm the absence of such a docket with high confidence — docket databases are not fully indexed by general web search — so treat "no CAFC 2026 activity found" as the search result, not a certified negative.

Bibliographic data

Abstract (as issued)

Video data defining a series of images is processed to define a first series of nodes in a first multidimensional space, each node corresponding to an image and positioned in dependence on features of that image. A transformation function maps each node in the first space onto a node in a second, lower-dimensionality space while maintaining neighborhood relationships between nodes. A second series of nodes is defined in the second space in accordance with the transformation function, and a clustering analysis is performed on the nodes of the second space.

Plain-language overview of the independent claims

There are two independent claims — claim 1 (method) and claim 12 (system) — and they are essentially parallel in scope. All other claims (2–11 and 13–22) depend from them.

Claim 1 — Method
A computer-implemented method for processing video (recited as configuring a digital data processor and memory) with four core steps plus a required sub-step:

1. Build the node set: place each image from the video as a "node" in a first high-dimensional space, where each node's position is set based on the appearance components of that image (i.e., visual apperance features).
2. Learn a transformation that preserves neighborhoods: define a function that maps those nodes into a second, lower-dimensional space, while keeping nodes that were near each other in the first space still near each other in the second space (a locality-preserving dimensionality reduction).
3. Map the nodes into the lower-dimensional space using that transformation.
4. Cluster the nodes in the lower-dimensional space to find clusters — these clusters represent significant/repetitive portions of the video.
5. Required sub-step (the narrowing limitation of claim 1): defining the transformation function must itself comprise (a) building a graph representing relationships between neighboring nodes of the first space, (b) assigning a weight to each such relationship, and (c) calculating the transformation function from that mapping (in the specification, this is the 2D-LPP generalized eigen-equation approach of equation (10)).

Claim 12 — System
The apparatus counterpart of claim 1: a video processing system with at least one processor and associated memory configured to perform the same four steps and the same graph-building/weighting/transformation-calculating sub-steps. Because claim 12 uses "configured for" language, it reads on a system programmed to carry out the method.

Notably, only claims 1 and 12 independently recite the graph/weight/transformation calculation. Omitting that sub-step would leave only the broader four-step abstraction pipeline, which is not claimed independently here.

Dependent claim landscape (brief)

• Claim 2/13: node positions defined by a pixel and/or feature mapping algorithm.
• Claim 3/14: one dimension of the first space represents time/sequence in the original video.
• Claims 4–6/15–17: graph definition via a predefined neighborhood condition (optionally set by preprocessing training data), with edge weights based on similarity.
• Claim 7/18: the transformation is computed from eigenvectors of the 2D-LPP generalized eigen-equation Aᵀ(L⊗I_m)Aw = λAᵀ(D⊗I_m)Aw, with A the (mN×n) column-stacked matrix, D the diagonal degree matrix, L = D − S the Laplacian, and ⊗ the Kronecker product.
• Claim 8/19: a second locality-preserving transformation into a third, even lower-dimensional space, with clustering performed there.
• Claim 9/20: outputting an image associated with an identified cluster.
• Claim 10/21: analyzing the video to identify shots and outputting the video of a shot associated with a cluster.
• Claims 11/22: user-selectable viewing of output images/shots (both depend respectively on claims 9 and 20).

Technical approach described in the specification

The patent applies 2D-LPP (two-dimensional locality preserving projection) to video frames. Video is first temporally and spatially sub-sampled (e.g., every 10th frame; reduced resolution such as QCIF or 80×60). Each frame is treated as a matrix node, a k-nearest-neighbor or ε-neighborhood content adjacency graph is built with an explicit temporal distance constraint (visually similar frames far apart in time are deliberately not linked, preserving video temporal structure), and similarity weights (binary or heat-kernel) are assigned. The transformation is learned by solving the generalized eigenvalue problem, and in preferred embodiments applied twice in tandem to reduce an m×n frame down to a d×d space (d ≤ 5, preferably ≤ 3). K-means (or a mixture-of-Gaussians) then clusters the reduced features; the frame nearest each cluster centroid becomes a storyboard key frame, subject to cluster-validity filtering.

Two worked examples are given in the specification:

• Example 1 (structured video): the "New Indians, Segment 01" documentary from The Open Video Project — 1,686 MPEG-1 frames at 320×240, temporally sub-sampled by 10 to 168 images, resized to 80×60, with k=3 and d=3, clustered into 20 key frames.
• Example 2 (home video): a 4-minute MPEG-1 clip at 25 fps, 5,995 frames of 352×288, sub-sampled to N=600 frames at 88×72, with k=3 and both d=3 and d=5, producing storyboards of size NC=20 and NC=8.

Uncertainty flags

• No CAFC 2026 docket or district-court litigation for US 8,311,394 was found in my searches. This is a search-result finding, not a definitive statement that none exists; litigation databases (PACER, Docket Alarm, Lex Machina) may not be fully surfaced by general web search.
• The only forward citation listed on Google Patents is US 11,205,103 B2 (a sentiment-analysis autoencoder Patent), which is a weak technical relationship and does not evidence commercial adoption.
• The pre-grant publication number is US 2009/0087122 A1 (the user-facing patent number is correctly US 8,311,394 B2; I have not altered any identifiers).

I'll search for litigation involving US patent 8311394 across patent litigation databases.

Let me dig deeper into the case that appeared and check Unified Patents and other litigation databases.

The Daingean case involves different patents (e.g., 8,576,803), not 8,311,394. Let me search more specifically via litigation databases.

Findings: Litigation involving US Patent 8,311,394

No known litigation involving US Patent No. 8,311,394 appears to exist, based on my searches of patent litigation resources (including Unified Patents-style litigation data surfaced through web results, CourtListener/RECAP records, PACER-derived materials, and RPX/legal-press coverage).

Patent identification (verified)

• Patent: US 8,311,394 B2 — "Video abstraction"
• Assignee: British Telecommunications plc (original and current)
• Inventors: Li-Qun Xu; Bin Luo
• Priority date: 2006-03-30 · Filed: 2007-02-21 (PCT/GB2007/000594) · Granted: 2012-11-13
• Family: US20090087122A1, EP1999753B1, WO2007113458A1, MY152525A

What I checked and excluded (to avoid false positives)

I want to be explicit that I did not conflate your patent with similarly numbered or same-party patents:

1. Daingean Technologies Ltd. cases (E.D. Tex., e.g. Daingean Technologies Ltd. v. T-Mobile USA, Inc., No. 2:23-cv-00347-JRG-RSP; and the AT&T case) — These assert patents from the Daingean/Atlantic IP portfolio: 8,576,803; 10,484,976; 10,841,958; 10,932,207; 11,134,400; 11,196,509. 8,311,394 is not among them. (Note: the "394" digits in Daingean's number are just the case-number suffix — unrelated to your patent.)

2. British Telecommunications PLC v. IAC/InterActiveCorp (D. Del., No. 1:18-cv-00366) — BT asserted a different set of patents: 6,578,079; 7,974,200; 7,243,105 (the sole surviving claim 10 of the '105 patent was held invalid under §101 in the Aug. 5, 2025 opinion). 8,311,394 is not among the asserted patents.

3. ASUS Technology Licensing / Innovative Sonic / Celerity IP v. AT&T, Verizon, T-Mobile (E.D. Tex., No. 2:23-cv-00486) — Quality-of-service / small-cell patents; not the '394 patent.

Important caveats

• I could not directly query the login-restricted Unified Patents litigation caselist or PACER's full docket system; my conclusion rests on the litigation data that surfaced through searches and public docket aggregators. It is possible a very recent, low-profile, or sealed case exists that did not surface.
• "No litigation found" is a negative finding, not a certification. A definitive clearance search would require a direct query of Docket Navigator, Lex Machina, or PACER's full-text search for "8,311,394" (and the EP family member EP1999753), which are paywalled/subscription tools.

Bottom line

I found no case — no plaintiff, defendant, jurisdiction, case number, filing date, or outcome — in which US 8,311,394 has been asserted. If you need a high-confidence negative, I recommend running "8,311,394" through Docket Navigator or Lex Machina directly, since those include sealed and very recent filings that public web results may omit.

I'll start by checking the canonical structured list (which reports no AIA trials) against live web sources, since the task requires flagging anything the ODP may not have indexed.

The initial searches surfaced IPRs against different patents (e.g., Maxell's 8,170,394 — a distinct patent). Let me target the exact number and possible owner/assertion channels.

No hits yet on the exact number. Let me probe a few more angles — the related EP family, potential assertion campaigns by BT, and the standard PTAB docket sources.

PTAB / AIA Trial Analysis — U.S. Patent No. 8,311,394 ("Video abstraction")

Verification date: 2026-09-28
Canonical source: USPTO Open Data Portal "PTAB proceedings on file" block supplied with this task
Supplemental source: live web search (PTAB public dockets, Google Patents, FreePatentsOnline)

Proceedings overview

Total AIA trial proceedings on file: 0 — the USPTO ODP structured block returns no Inter Partes Review, Post-Grant Review, or Covered Business Method proceeding for US 8,311,394, and my independent web search surfaced no petition, institution decision, Final Written Decision, or appeal naming this patent. Breakdown by status therefore stands at 0 active / 0 claims invalidated / 0 claims sustained / 0 settled / 0 institution denied.

The defensive posture this gives a defendant is the opposite of a hardened patent and the opposite of a dead one: 8,311,394 is an untested patent. No claim has been canceled, none has been confirmed, and no petitioner — not even a defensive aggregator — has put the 2D-LPP video-abstraction claims in front of the Board. A defendant cannot point to an FWD and say "claim 1 is already dead," but equally the patent owner cannot say "this patent survived two IPRs." Every validity ground is still live and un-estopped; the invalidity work, if you want it done, starts from scratch.

Proceedings on file

None. Because the Board has never instituted a trial on this patent, the per-proceeding template (type / filed / status / panel / grounds / institution / FWD / settlement / appeal / defensive value) has no content to populate. I am not filling it with speculation.

⚠️ Near-miss dockets — do NOT confuse these with 8,311,394

A defendant reading PTAB dockets quickly will see "'394" shorthand and assume this patent has been litigated at the Board. It has not. The following proceedings involve different patents and are a known trap:

None of these share a specification, a priority chain, or an inventor with US 8,311,394.

Cross-checking the absence

I applied three independent corroborating checks:

1. Google Patents "Cited By" for US8311394 lists a single forward citation — US11205103B2 (The Research Foundation for the State University of New York, priority 2016-12-09, "Semisupervised autoencoder for sentiment analysis"). That reference is an unrelated machine-learning patent and is not a PTAB proceeding. No IPR/PGR/CBM decision is listed as citing this patent.
2. Family / parallel-prosecution check. The family is US 8,311,394 (granted 2012-11-13; adjusted expiration 2029-10-08), EP1999753B1 (granted 2018-01-10), MY152525A, and WO2007113458A1; the sibling EP2224357A1 ("Video segmentation," BT) is a separate case. No US trial or parallel validity proceeding appears for any of these.
3. Assignee / assertion check. Current assignee is British Telecommunications plc; inventors Li-Qun Xu and Bin Luo (assignment recorded 2008-09-29, REEL/FRAME 021599/0930). BT's US patent-enforcement activity in this period (e.g., the Vimeo/IAC and Match.com matters, and the Fastnets streaming patents) involved different patents in BT's portfolio, not the '394 video-abstraction patent. I found no BT assertion campaign and no defensive aggregator (e.g., Unified Patents) IPR tied to 8,311,394.

Confidence / caveat. I state this with moderate-to-high confidence, not certainty: the ODP block is authoritative as of its ingest date, and my searches found nothing contradicting it, but a petition filed very recently could exist in PTAB E2E / PatentCenter before it propagates to third-party dockets. Confirm live before relying on it (see next steps).

Strategic summary

Claim status: all 22 claims UNTESTED. US 8,311,394 has two independent claims — claim 1 (method) and claim 12 (system) — with dependent claims 2–11 and 13–22. Not one of them has been canceled, confirmed, or amended through an AIA trial. There is no surviving-claim list to report because there is no adjudication to narrow from; the full claim set as granted is what a defendant must contend with, subject only to whatever the courts might do.

Estoppel landscape: nothing has attached. Because no IPR has reached a Final Written Decision, § 315(e)(2) estoppel is inapplicable to anyone. That is a double-edged sword for a defendant. On the one hand, you are not riding anyone else's petition and cannot lean on another party's invalidity win. On the other hand, you are also not barred — you (or a co-defendant, or a defensive aggregator) can file a first IPR on any § 102/§ 103 ground, including art the patent owner has never seen, without § 325(d) "already before the Office" squabbling beyond the original prosecution. The absence of a prior petitioner also means no General Plastic / follow-on-petition discretionary-denial risk against your first filing. If you are sued and want a PTAB path, the field is open — but note the § 315(b) one-year bar from service of an infringement complaint.

Pattern signals: none. There is no serial-petitioner pattern (no single entity has filed even one IPR), no patent-owner appeal history to the Federal Circuit on this patent, and no defensive aggregator in the chain. That combination — a twelve-year-old, still-in-force BT patent in a commercially hot area (video summarization, automatic storyboarding, home-video abstraction) with zero PTAB challenges — is unusual. It most plausibly reflects limited US assertion of this particular patent rather than a robust validity shield: well-asserted patents in crowded art tend to attract IPRs. Do not read the zero as "this patent is bulletproof"; read it as "nobody has yet bothered to test it."

Recommended next steps

1. Confirm the zero live. Before advising a client, pull the "PTAB proceedings" / "AIA trials" tab for US 8,311,394 on USPTO PatentCenter and the PTAB E2E / PTAB public-information portal (https://ptacts.uspto.gov/ptacts/), and search the Board docket by patent number (not by the "'394" shorthand — see the near-miss table above). Do the same on CourtListener / Docket Alarm for any CAFC appeal captioned with this patent.
2. There are no FWDs, institution decisions, or appeals to link or quote. I have deliberately not cited any FWD for this patent because none exists; citing one would be fabrication. If you need a model of how a video-summarization claim set fares at the Board, the IPRs in the near-miss table are instructive but not precedential for 8,311,394.
3. If you are a defendant facing assertion: the strategic value here is initiative, not leverage. Because no estoppel has run, you are free to build a first-challenge record — but respect the § 315(b) one-year clock from service of the complaint, and remember § 311(b) limits IPR grounds to patents and printed publications. The specification's own admission that the core technique is 2D-LPP (Chen et al., reference [3]) and that LPP itself is prior work by He et al. (references [6] and [7]) is a useful starting point for a § 103 obviousness theory, and the file history will show whether 2D-LPP's application to video (vs. face recognition) drew any specific prior-art rejection.
4. If you are the patent owner: the clean PTAB record is a genuine asset for licensing — no adverse FWD and no narrowing amendment to explain — but it is fragile. A single successful IPR would remove that asset entirely, and the 2D-LPP/LPP art cited in the specification makes the claims a plausible target. Consider whether the strongest independent claims (1 and 12) are worth reinforcing through reissue, continuation, or targeted licensing before a challenge materializes.

Bottom line for counsel: No PTAB activity on file for US 8,311,394 as of 2026-09-28. The patent's 22 claims stand untested and un-estopped; proceed on the assumption that any AIA challenge would be a first-of-its-kind filing, and verify the docket live before relying on the absence in a brief.

Ownership & Assignment Analysis — US 8,311,394 B2 ("Video abstraction")

Scope note / sourcing limitation. The USPTO Assignment Center (assignmentcenter.uspto.gov) and its legacy mirror (assignment.uspto.gov) are not directly retrievable by the search tooling available for this task, and the specific attorney "correspondent of record" field is not exposed in any source I could reach. My assignment data is therefore drawn from the Google Patents legal-events block for US 8,311,394 (mirrored from USPTO assignment records and carrying the reel/frame number), corroborated by the FreePatentsOnline (FPO) face-of-patent record and the family/priority data in the patent text. Where a datum (notably the recording correspondent) is not exposed, I say so rather than invent it. Verify against the Assignment Center search page: https://assignmentcenter.uspto.gov/ (search by patent number 8311394).

Inventors

Unusual-pattern check (inventor departures within 12 months of filing): Not observed / insufficient data. A departure of an inventor from the original assignee can only be inferred from a subsequent assignment of the patent or a public employment change. There is no recorded post-filing assignment of this patent away from BT, and I found no public record of either inventor's separation from BT. I cannot certify a negative here.

Original assignee

British Telecommunications Public Limited Company (the operating company of the BT Group; issuer: "British Telecommunications PLC" per Google Patents, "British Telecommunications plc" on the FPO face-of-patent record). The pre-grant publication US 2009/0087122 A1 lists the inventors as applicant-assignees bridging to BT.

• Primary line of business: incumbent UK fixed/mobile telecommunications and network services — telephony, broadband, IP/transport networks, pay-TV (BT TV), conferencing, and video/network research. This is a large operating company, not a holding vehicle.
• Does it ship products/services embodying the claims? BT is unquestionably a commercial operator of video and network platforms, and its research organisation produced the video-abstraction prototype described in the specification (a "prototype video summary system for flexible and interactive user-oriented video browsing," per the companion CIVR '07 paper). Whether any shipped BT product practices the specific claimed 2D‑LPP pipeline is not established — the specification describes a research prototype, not a commercial SKU. Treat "embodies the claims" as unproven.
• Current status: Operating. BT is a solvent, publicly listed telecom group; it has been an active patent plaintiff in its own right (e.g., British Telecommunications plc v. Prodigy Communications Corp., S.D.N.Y. 7:00‑cv‑09451, 189 F. Supp. 2d 101 (2002), over the "Sargent patent" US 4,873,662; and the BT/ASSIA cross-licence and infringement disputes litigated in the UK, including Adaptive Spectrum and Signal Alignment Inc v British Telecommunications plc [2014] EWHC 2730 (Pat) and the 2022–2023 follow-on licence-construction appeals). None of those proceedings involved US 8,311,394. No bankruptcy, dissolution, or acquisition of BT's patent estate.

Assignment timeline

The full recorded chain is one entry:

• 2007-03-12 to 2007-03-21 (executed — "SIGNING DATES FROM 20070312 TO 20070321") / recorded 2008-09-29 — Reel 021599 / Frame 0930
  • Conveyance: Assignment ("ASSIGNMENT OF ASSIGNORS INTEREST")
  • Assignor: XU, LI-QUN; LUO, BIN (the two named inventors)
  • Assignee: BRITISH TELECOMMUNICATIONS PUBLIC LIMITED COMPANY
  • Correspondent: Not exposed in the sources retrieved. The Google Patents legal-events text records only the assignors, assignee, execution dates, and reel/frame; it does not surface the recording correspondent field. The prosecution attorney/agent of record printed on the face of the patent is NIXON & VANDERHYE, PC (per FreePatentsOnline) — that is the prosecuting firm, not evidence of the assignment-recording correspondent, so it should not be conflated. Flagged as unavailable, not inferred.
  • Context: Routine employee/inventor-to-company assignment — the standard pre-issuance perfection of title from the inventors to the corporate applicant, recorded during national-phase entry (US application 12/295,033 entered the national phase from PCT/GB2007/000594).

No post-issuance assignments are recorded. The Google Patents legal-events block contains no assignment after the 2008-09-29 inventor→BT record. All subsequent entries are administrative: patent-grant status (2012-10-24), fee-payment/entity-status procedural entry (2015-12-04), and maintenance-fee payments at the 4th year (2016-05-05), 8th year (2020-04-22), and 12th year (2024-04-18), all recorded against the same large-entity owner. Continued payment of the 12th-year fee is consistent with BT still holding title and valuing the asset.

Ownership-chain gaps to be aware of: a single recorded assignment does not prove that no unrecorded transfer occurred. However, there is no evidence of any transfer to a licensing entity, no security agreement, no merger/change-of-name record, and no release anywhere in the record — the absence of a transfer is affirmative evidence of retention here because the maintenance fees continue to be paid under the original owner's entity profile.

Timeline diagram

timeline
    title Ownership of US 8311394
    2006 : EP priority application filed
    2007 : PCT application filed 21 Feb
         : Inventors execute assignment to BT
    2008 : Inventor to BT assignment recorded USPTO
         : Reel 021599 frame 0930
    2012 : US patent issued 13 Nov
    2024 : 12th year maintenance fee paid

NPE / troll-pattern signals

Verdict

Insufficient data — per the rubric's explicit "only the original assignment" condition.

Justification: The record contains exactly one assignment — inventors Xu and Luo to British Telecommunications Public Limited Company, executed 2007-03-12→2007-03-21 and recorded 2008-09-29 at reel 021599/0930 — and nothing thereafter; all later entries are fee payments (4th/8th/12th year) and a grant-status flag, which indicate the original owner retained and maintained the patent. Not a single NPE-tell factory signal is present: no shell transferee, no known asserter, no cascading transfers, no bankruptcy, no defensive aggregator. This is the ordinary profile of a large operating-company patent that was never monetised or divested; the rubric's more affirmative "Operating-company assertion" verdict is unavailable because there is no evidence BT shipped a product practicing these claims and no evidence of any suit on this patent. One caveat cuts the other way and should be stated openly: because BT is a proven, willing patent plaintiff (Prodigy; ASSIA), the current owner is an operating company with the capacity to assert — but capacity is not evidence, and the honest call on the record is Insufficient data / no NPE signal.

Verification links:

• USPTO Assignment Center search: https://assignmentcenter.uspto.gov/ (search "8311394"; legacy mirror: https://assignment.uspto.gov/patent/index.html)
• Google Patents legal events (source of reel 021599/0930): https://patents.google.com/patent/US8311394/en
• Family/priority (EP 1999753 B1; WO 2007/113458 A1; MY 152525 A): same Google Patents page, "Also Published As" section.

I'll search for authoritative information on this patent and its cited prior art.

Let me verify the prior art citations and their family relationships, and check the USPTO record.

Let me verify the NEC prior-art family and check the search-report citations for the family.

Prior-art analysis for US 8,311,394 B2 ("Video abstraction")

Method note / source caveat. I was not able to open USPTO PatentCenter, Patent Full‑Text (PatFT), or the Patent Assignment Search directly in this session; my tool is general web search. The citation list below is therefore taken from the USPTO-record mirrors that reproduce the examiner's references (PTO‑892) — Google Patents' "Patent Citations" table for the patent and the FreePatentsOnline copy of the same record. If you need a certified PTO‑892 (form SB/08a/892), that must be pulled from the file wrapper in PatentCenter for application 12/295,033. I flag this because the two mirrors agree with each other, but neither is the primary source.

Anchor. The record is for the correct number: US 8,311,394 B2, application 12/295,033, "Video abstraction," British Telecommunications PLC, inventors Li‑Qun Xu and Bin Luo. This analysis uses the issued claims (1–22) as set out in the authoritative text supplied.

1. The complete patent-citation list for US 8,311,394

Google Patents lists three documents under "Patent Citations," and one forward citation under "Cited By":

That is the entire patent-citation set. Only one of the three "citations" is actually prior art. The other two are the applicant's own documents — the PCT publication and the US pre‑grant publication of the same application — which the examiner had to list on the face of the patent but which cannot be used to reject it. I treat them separately below.

2. US 2004/0170321 A1 — the only substantive patent reference

Full citation: US 2004/0170321 A1, "Method and system for segmentation, classification, and summarization of video images," NEC Corporation; appl. no. 10/691,222; filed Oct. 20–21, 2003; earliest priority Nov. 24, 1999; published Sep. 2, 2004. (Sources conflict slightly on the filing day — 2003‑10‑20 vs. 2003‑10‑21 — and on the priority day, 1999‑11‑23 vs. 1999‑11‑24; I have not auto-corrected these; the substantive point is unaffected.)

Family (relevant because the same disclosure also exists as grants):

• US 09/556,349 → US 7,016,540 B1 (granted 2006‑03‑21), priority 1999‑11‑24.
• US 10/691,222 → US 7,151,852 B2 — the grant corresponding to the cited publication.
• US 11/505,833 (filed 2006‑08‑18, a division of 10/691,222) → US 7,630,562 B2 (granted 2009‑12‑08).
• Continuation publication US 2006/0280365 A1 (published 2006‑12‑14).

Brief description. The NEC disclosure is an SVD-based video segmentation/summarization system. Each frame is represented by a vector of concatenated colour-histogram descriptors; a singular value decomposition maps those vectors into a "refined feature space" of fewer dimensions (the reference expressly frames SVD as "deriving a low dimensional refined feature space from a high dimensional raw feature space … and capturing the essential structure of a data set"). In that reduced space a content/value metric is computed per shot; the most static shot is defined as an "information unit" and its content value is used as a threshold to cluster the remaining frames. Output is either an optimal set of static key frames or a summarised motion video. Frame similarity/segmentation is done by distance in the refined feature space.

§ 102 assessment (claim-by-claim).

• Qualifies as prior art. Published 2004‑09‑02, i.e. more than one year before the 2006‑03‑30 priority date, so it is § 102(b) art (pre‑AIA, which governs here) as well as § 102(e) art via its earlier US filings.
• Claim 1 (independent method): NOT anticipated. NEC does disclose a rough structural analogue of steps 1, 2 and 4 — frames positioned in a high-dimensional space, mapped into a lower-dimensional space, then clustered, with images/shots selected from clusters. But claim 1's closing "wherein" clause requires the transformation to be defined by (a) building a graph of neighbouring nodes, (b) weighting each relationship, and (c) calculating the transformation from that mapping. NEC's SVD is a global, variance-maximising decomposition; it builds no neighbour graph, assigns no edge weights, and solves no generalized eigen-equation. SVD does not preserve neighbourhood relationships in the locality-preserving sense claim 1 requires. On its face, then, NEC does not disclose every element of claim 1 as arranged in the claim, so it cannot anticipate it.
• Claim 12 (independent system): NOT anticipated for the identical reason — claim 12 recites the same graph/weight/transformation-calculating sub-steps.
• Dependent claims 2–11 and 13–22: NOT anticipated as claims, because each depends from and incorporates all limitations of claim 1 or 12. Two of them nonetheless map onto NEC's disclosure at the element level and would be the parts of NEC an examiner would lean on:
  • Claim 9 / claim 20 ("identifying an image … associated with that cluster and outputting that image") — NEC outputs key frames per cluster.
  • Claim 10 / claim 21 (identify shots, output the video of a shot associated with a cluster) — NEC segments into shots and outputs a summarised motion video.
  • Claim 2 / 13 (node position from pixel/feature mapping) — NEC's histogram-vector frame representation.

Bottom line on US 2004/0170321 A1: it is the closest patent reference of record and is squarely in the same field (video abstraction/summarisation), so it is highly usable as § 103 obviousness art and as evidence of the state of the art — but it does not meet the graph/weight/eigen-decomposition narrowing that is the actual point of novelty in independent claims 1 and 12, and so it does not anticipate any claim of US 8,311,394 as issued. Treating it as a § 102 reference would be an over-read.

3. WO 2007/113458 A1 — not prior art

Full citation: WO 2007/113458 A1, "Video abstraction," British Telecommunications Public Limited Company, priority 2006‑03‑30, published 2007‑10‑11. This is PCT/GB2007/000594 — the very international application of which US 12/295,033 is the US national phase (the priority application is EP 06251783.4).

Why it is not prior art, and § 102 position. It is the applicant's own earlier publication of the same invention by the same inventive entity. It does not pre-date the 2006‑03‑30 priority date, and even if it did, a party's own disclosure of the same subject matter cannot be used against its own later application in this posture. It cannot anticipate or render obvious any of claims 1–22. Its appearance on the face of the patent is an artifact of family listing, not a substantive rejection.

4. US 2009/0087122 A1 — not prior art

Full citation: US 2009/0087122 A1, "Video abstraction," Li‑Qun Xu (etal.), published 2009‑04‑02. This is the pre‑grant publication of the instant application itself (US 12/295,033, filed as the national phase of PCT/GB2007/000594), i.e. the publication that later issued as US 8,311,394 B2.

Why it is not prior art, and § 102 position. A reference cannot be prior art against the very application from which it sprang. The disclosure is identical and the effective date is the same 2006‑03‑30 priority date, so it is neither § 102(a)/(b) (not before the invention date/priority date) nor § 102(e) art against its own case. It cannot anticipate or render obvious any claim 1–22.

5. What arguably is the most technically relevant prior art — the non‑patent citations

Because only one patent citation is substantive, the real prior-art pressure on this patent comes from the 34 non‑patent references of record. The most technically relevant are:

• Chen, Zhao, Kong & Luo, "Two‑Dimensional LPP for Image Recognition," Proc. IEEE 2005 Int'l Conf. on Intelligent Computing, Aug. 23–26, 2005, pp. 1159–1168 — reference [3] of the specification. This discloses the 2D‑LPP method itself, including the graph construction and the generalized eigen‑equation that claim 7 / claim 18 recites nearly verbatim. Predates the 2006‑03‑30 priority date. This is the single most relevant reference for the transformation limitation; note, however, that it is directed to image (face) recognition, not video abstraction, so it does not alone anticipate claims 1/12.
• He & Niyogi, "Locality preserving projections" (NIPS 16, 2004) — reference [6]; and He, Yan, Hu, Niyogi & Zhang, "Face recognition using Laplacianfaces," IEEE TPAMI 27(3):328–340, 2005 — reference [7]. The underlying LPP method.
• Hanjalic & Zhang, "An integrated scheme for automated video abstraction based on unsupervised cluster‑validity analysis," IEEE TCSVT 9(8):1280–1289, Dec. 1999 — reference [5]. The video‑abstraction‑by‑clustering‑with‑validity‑filtering framework that the patent's claim 8/19 and cluster‑validation passages build on.
• Kokiopoulou & Saad, "Orthogonal Neighborhood Preserving Projections," 5th IEEE ICDM, pp. 234–241, Nov. 27, 2005 — a neighbourhood‑preserving projection alternative cited on the face of the patent; relevant § 103 art to the "preserve neighbourhood relationships" limitation.
• Graves et al., "Surveillance video indexing with iconic patterns of activity," IEE VIE, Apr. 6, 2005, pp. 409–416 — video indexing by learned patterns.

The most likely § 103 combination against independent claim 1 would be one of the video‑abstraction references (e.g. Hanjalic & Zhang, or the NEC publication) as the video‑summarisation framework combined with the 2D‑LPP teaching of Chen et al. (and/or He et al.) for the graph‑weighted, neighbourhood‑preserving eigen‑projection — with the inventors' own CIVR '07 paper, Xu & Luo, "Appearance‑based Video Clustering in 2D Locality Preserving Projection Subspace," pp. 356–363 (Jul. 9–11, 2007), noted as the inventors' publication post‑dating the priority date and therefore not itself available as prior art.

6. Flags and uncertainties

• Only one of the three patent citations (US 2004/0170321 A1) is prior art; the other two are the applicant's own family documents. Any statement that "three patent references were cited against this patent" would mischaracterise the record.
• No patent citation of record anticipates any claim of US 8,311,394. The NEV reference fails the graph/weight/eigen‑equation narrowing of claims 1 and 12; claims 2–22 each incorporate that narrowing via dependency.
• Discrepancies in the NEC family dates (filing 2003‑10‑20 vs. 2003‑10‑21; priority 1999‑11‑23 vs. 1999‑11‑24) are left as‑found; they do not affect the § 102(b) conclusion (the 2004‑09‑02 publication pre‑dates the 2006 priority by well over a year).
• I could not independently confirm the PTO‑892 directly (no direct PatentCenter/PatFT access in this session); the citation set is from USPTO‑record mirrors that agree with one another.
• Date note: the task header says April 26, 2026; the authoritative fetch stamp on the full text is 2026‑09‑28. This does not change any substantive conclusion.
• No litigation, PTAB, or CAFC 2026 docket for US 8,311,394 was found in my searches, consistent with the earlier summary; this remains a search‑result finding, not a certified negative.

Grounding sources: Google Patents record — https://patents.google.com/patent/US8311394/en ; FreePatentsOnline mirror of the same record — https://www.freepatentsonline.com/[8311394](/patent/8311394).html ; NEC cited publication — https://portal.unifiedpatents.com/patents/patent/US-20040170321-A1 and https://patents.google.com/patent/US20040170321A1/en ; NEC family grant US 7,630,562 — https://patents.google.com/patent/[US7630562B2](/patent/US7630562B2) .

I'll ground my obviousness analysis by verifying the key prior-art references identified on the patent's page. Let me search for the most probative ones.

Obviousness Analysis — US 8,311,394 B2 under 35 U.S.C. § 103

Scope note. This analysis builds on the bibliographic and claim material in the prior sections. I used the prior-art references listed in the "Patent Citations" and "Non-Patent Citations" sections of the patent page itself (as instructed), and confirmed the existence, dates, and content of the most probative ones by search. I have not altered any patent number or identifier. Where a conclusion depends on a reference I could not read in full, I say so.

1. Legal framework and level of ordinary skill

The application was filed 2007-02-21 (PCT national phase) with a 2006-03-30 EP priority, so pre-AIA §§ 102/103 govern. Every reference relied on below published before 2006 and therefore qualifies as § 102(b) prior art. Obviousness is assessed under Graham v. John Deere and KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007): the scope/content of the prior art, the differences from the claims, the PHOSITA level, and objective indicia. Under KSR, a combination need not be taught explicitly; a reasoned motivation, a predictable result, or the mere "combination of familiar elements according to known methods" suffices.

PHOSITA (as of March 2006): a person with an M.S./Ph.D. in EE or CS (or a bachelor's plus ~3–5 years' experience) in multimedia information management, with working knowledge of (i) content-based video retrieval and video abstraction/segmentation, (ii) appearance-based image representation (color histograms, pixel matrices), and (iii) dimensionality reduction and clustering (PCA, LDA, LPP, spectral methods, k-means). This is exactly the population that authored the cited literature (He/Niyogi, Yang/Zhang, Chen et al., Zheng et al., Hanjalic/Zhang).

2. The prior-art arsenal (from the patent's own citation lists)

[NEC] is the only patent reference of record; [Chen], [He], [Zheng], [HZ] are in the IDS/Non-Patent Citations. The examiner allowed the claims over [NEC] alone, which confirms [NEC] does not teach the graph/weight/eigenvector projection. The obviousness case therefore turns on joining [NEC] to the 2D-LPP literature.

3. Element-by-element mapping of independent claim 1

Every limitation of claim 1 has an express counterpart; limitation (e) — the narrowing sub-step — is met essentially word-for-word by [Chen]'s three-stage algorithm.

4. Primary combination and motivation

Combination A (strongest): [NEC] + [Chen], optionally + [HZ] or [Zheng]

Why a PHOSITA would combine (KSR-compliant rationales):

1. Same field of endeavor. Both [NEC] and [Chen] address appearance-based representation and analysis of image/video data; [NEC] explicitly targets video summary and [Chen] targets image recognition using the same matrix-of-pixels paradigm.
2. Explicit motivation in the lead reference. [NEC] itself performs dimensionality reduction (SVD) before clustering precisely because raw color-histogram vectors are high-dimensional ("deriving a low dimensional refined feature space from a high dimensional raw feature space"). The patent's own Background concedes the "curse of dimensionality" and that "dimensionality reduction techniques" are a known solution (its refs [2], [16]). A PHOSITA is thus affirmatively directed to improve the DR step.
3. Known, better DR methods available. LPP, 2D-PCA, 2D-LDA and 2D-LPP were published 2004–2005 by the same research community [Chen], [He], [Yang], [Li-Yuan], [Kokiopoulou]. [Chen] demonstrated 2D-LPP outperformed 2D-PCA and 2D-LDA. Substituting a locality-preserving projection for SVD/PCA is the kind of predictable substitution KSR holds obvious ("if a technique has been used to improve one device … using the technique is obvious").
4. The clustering step is already conventional in both worlds. [HZ] and [Zheng] teach clustering video frames/images and selecting centroids as key items; [NEC] clusters in its refined space. The combination merely reorders known steps.
5. Reasonable expectation of success. A linear projection followed by k-means is a fully predictable pipeline; [Zheng] empirically shows the LPP-then-k-means combination beats "PCA + Kmeans" on image data.

Result: claim 1 (and claim 12, its apparatus twin, which differs only in "configured for" language) is rendered obvious.

Combination B: [HZ] + [Chen] (+ [He])

[HZ] supplies the video-abstraction-by-clustering framework (including unsupervised cluster-validity selection, a feature the patent adopts in its block 11). [Chen]/[He] supply the graph-based locality-preserving projection that replaces the unspecified feature representation. [HZ]'s key-frame-as-cluster-centroid step likewise anticipates the patent's centroid-selection step. Motivation is again the dimensionality-reduction rationale.

Combination C: [Zheng] + [NEC] (or [HZ])

[Zheng] is arguably the closest single teaching of the core pipeline: it treats LPP as a projection and runs k-means in the LPP subspace, and it is applied to images. Extending an image-clustering method to video frames (themselves images) requires only the recognition that video is a time-ordered image sequence — which [NEC], [HZ], and [Sun] all supply. Adding 2-D LPP ([Chen]) addresses the high per-frame dimension.

5. Dependent claims

• Claims 2/13 (pixel/feature mapping): met by histogram vectors [NEC] and pixel matrices [Chen]/[Yang].
• Claims 4–6/15–17 (graph via neighborhood condition; training-data preprocessing; similarity edge weights): met by [Chen] (k-NN/ε graph; binary/heat-kernel weights) and [He].
• Claim 7/18 (the specific eigen-equation Aᵀ(L⊗I_m)Aw = λAᵀ(D⊗I_m)Aw): met verbatim by [Chen]. This is the single most damaging reference, because the claim language and the reference equation are identical.
• Claim 8/19 (a second neighborhood-preserving projection to a third space): supported by the art's routine use of cascaded projections — [He]'s PCA-before-LPP (Laplacianfaces, discussed in the patent's own Background) and [Chen]'s tested "2D-LPP plus PCA." A second application of the same locality-preserving operator is a predictable extension; note, however, that claim 8 requires the second transform to preserve neighborhoods (not merely PCA), so this claim is the weakest link in a pure [NEC] + [Chen] theory. It is best attacked with [Kokiopoulou] (a second NP-preserving projection) or by arguing iterative 2-D LPP is a design choice.
• Claims 9/20 (output an image per cluster): met by [HZ] (centroid key frames) and [NEC].
• Claims 10/21 (identify shots; output the shot's video): met by [NEC] (shot segmentation + summary video) and [HZ] (preview by concatenating shots).
• Claims 11/22 (user-selectable viewing): met by [NEC]'s stated support for "different user requirements for video browsing and content overview."

6. Honest weaknesses in the obviousness case (and how they cut)

1. Claim 3/14 — "one dimension … represents the time or sequence." None of the DR references ([Chen], [He], [Yang]) positions images along a time axis; they operate on unordered image sets. This is the element with the least direct support. Counter: [NEC] segments along the time axis, and its own Background cites Yeung et al. using hierarchical clustering on keyframes "based on their visual similarity and temporal distance." So temporal information as a feature/clustering dimension is in the art, but the mapping to claim 3 is inferential, not literal.
2. Specification's temporal-distance edge constraint (block 4) — the visually-similar-but-temporally-distant frames are deliberately not linked. This is the patent's most inventive-sounding idea. Critically, it is not recited in claim 1 (claim 1's graph step says only "neighboring nodes of the first space"). If the applicant ever sought to rely on it, it lives only in the written description (and, partially, in claim 3). For validity of the issued claims as written, this does not rescue claim 1.
3. Claim 8's two-stage locality-preserving cascade (see §5) — the weakest dependent claim.
4. Objective indicia. The record on the patent page shows no evidence of commercial success, licensing, copying, or industry praise, and no secondary-considerations evidence at all. The patent's own results are framed as "comparable" to / "very similar" to manual storyboards — i.e., expected quality rather than an unexpected technical result. The noted omissions (frames 379/430/1358; the conservatory-base and white-van scenes) are explained by the inherent behavior of the projection under fast camera motion, which cuts against any "unexpected results" argument.
5. The inventors' own CIVR '07 paper ("Appearance-based Video Clustering in 2D Locality Preserving Projection Subspace") post-dates the 2006-03-30 priority and is therefore not prior art; I flag it only because it appears in the IDS and shows the technique was part of the community's ordinary toolkit. I did not treat it as § 103 art.

7. Bottom line

• Claim 1 and claim 12 are, in my assessment, obvious over [NEC] in view of [Chen] (with [HZ] and/or [Zheng] as reinforcing secondary references). The motivation is supplied by [NEC]'s own use of dimensionality reduction before clustering, by the patent's Background admission that high-dimensional clustering is a known problem solved by known DR methods, and by the 2004–2005 2D-projection literature. Claim 7's equation is matched verbatim by [Chen].
• Claims 2, 4–7, 9–22 tracks are obvious on the same or additional art as mapped above.
• The least-obvious claims are 3/14 (time dimension) and especially claim 8/19 (second neighborhood-preserving projection); these would be the natural focus of any validity challenge and are where a patentee's rebuttal would be strongest.
• No secondary-considerations evidence appears in the record, so there is nothing to weigh against the prima facie case.

Caveats on confidence: I read [NEC] through its abstract/claims and search snippets, and [Chen] through its abstract and equation excerpts (search results), not the full texts of each; my mappings of those two references are grounded in that retrieved content. [HZ] and [Zheng] abstracts were confirmed via ACM DL. The negative on objective indicia rests on the patent page, which is a limited record; a complete file-history and commercial-evidence review could change the Graham factor analysis, though it would not change the element-by-element mappings.