US 7583294

I'll search for authoritative information on this patent, including any 2026 litigation activity.

Let me dig into the claims and any litigation/appeal activity.

Let me retrieve the actual claim text and check for any 2026 appellate activity.

US 7,583,294 B2 — "Face detecting camera and method"

Scope note up front: I could not directly query the USPTO PatentCenter or the CAFC docket system from this session; the findings below come from the authoritative full patent text you supplied, plus Google Patents, Justia, Unified Patents, and PTAB/IPR tracking pages. I found no evidence of any 2026 CAFC docket activity for this patent. I flag several 1-day date discrepancies between sources below rather than reconciling them.

1. Bibliographic data

Unrelated-ID warning: A search for the bare string "7583294" surfaced JP 7583294 B2 (Japanese Patent No. 7583294, registered 2024-11-06, Shinagawa Refractories — refractory brick raw-material composition). That is an entirely different, unrelated document and has nothing to do with this analysis. I am reporting it only so it is not mistaken for the U.S. patent.

2. Abstract (verbatim)

"A method for determining the presence of a face from image data utilizes at least first and second algorithms. The first algorithm prescreens the image data, by determining a plurality of face candidates utilizing a pattern matching technique that identifies image windows likely to contain faces based on color and shape information. The second algorithm processes the face candidates determined by the first algorithm, and uses a posterior probability function classifier to determine the presence of the face."

3. Overview of the disclosure

The patent is a two-stage, resource-constrained face detection system intended to run inside a digital camera (CPU 30, RAM 42/ROM 44, "less than one second" target). The two detectors are named in the specification:

• Component W — a fast pre-screener derived from Wu, Chen & Yachida, "Face Detection from Color Images Using a Fuzzy Pattern Matching Method," IEEE TPAMI 21(6), 557–563 (1999). It uses skin/hair probability tables in an (L,s,t) color metric plus m×n (m=12, n=10) "shape models" for facial poses, with a sigmoidal non-linearity mapping probabilities to occupancy fractions and a "fuzzy relation" similarity score. It is tuned (parameters a and b set low) so it almost never misses a face, at the cost of many false positives.
• Component S — a low-false-positive Bayesian pattern-matching classifier based on Schneiderman & Kanade, "Probabilistic Modeling of Local Appearance and Spatial Relationships for Object Recognition," Proc. CVPR 1998, 45–51. It decomposes a standardized 56×56 face window into overlapping 16×16 subregions, applies PCA to 12 dimensions, sparse-codes to 6 coefficients, Lloyd-Max quantizes them (~1,000,000 "q1" values), handles positional dependence via vector-quantized distributions, and runs at multiple resolutions (nmags).

The combined detections are merged (the spec notes averaging overlapping eye locations outperformed picking the strongest detection). Face findings then feed camera functions: autofocus (measuring frame 49a), auto-exposure, flash determination, composition advice (rule of thirds, subject size, border clipping, varied head heights), color balance weighted to skin, red-eye detection/correction corroborated by face presence, orientation tagging, and face labeling/album generation.

4. Plain-language overview of the independent claims

Claim 1 — Method (two-stage face detection).
Covers the core method: (a) prescreen the image data with a first algorithm that produces a set of face candidates using a pattern-matching technique that looks for image windows likely to contain faces based on color and shape information; then (b) operate on those candidates with a second algorithm that applies a posterior probability function classifier to decide whether a face is actually present. In plain terms: a cheap color/shape "triage" step followed by an expensive statistical "confirmation" step.

Claim 5 — Method (the fast/accurate division of labor).
A method claim of the same two-stage family that expressly recites the performance asymmetry: the first algorithm has a first (higher) rate of false positives, and the second algorithm has a second (lower) rate of false positives than the first. This is the "illustrative embodiment" language quoted in the Summary of the Invention, and it is the practical heart of the invention — deliberately letting the prescreener over-detect so the accurate classifier only has to run on a few regions. Claims 10–18 depend from claim 5, including the camera-implemented variants (claims 6, 13, 14) that recite a capture section, electronic processing section, face data means, storage medium, recording means, and (in claim 13) an orientation algorithm; claim 14 adds a red-eye detection algorithm and display warning; claims 16–17 add red-eye corroboration by face presence.

Claim 19 — Computer program product.
An independent claim to a computer-readable storage medium storing a program that performs steps (a) and (b) of the two-stage method, and that includes the restriction that the first algorithm has a first false-positive rate and the second algorithm has a lower false-positive rate. In plain terms: the same fast-screener-then-accurate-classifier idea, claimed as software on a medium rather than as a process.

Drafting note: claims 6, 13 and 14 are drafted as dependent method claims ("The method as claimed in claim 5 wherein the method is implemented in a digital camera …") that nevertheless recite apparatus structure. Independent claims are 1, 5, and 19; total claim count is 19.

Uncertainty flag: I retrieved only the partial claim text (claims 6–19, plus the claim 1/5/19 gist) from the secondary source (Justia, https://patents.justia.com/patent/7583294). The exact wording of claims 1–5 is inferred from the abstract, the Summary, and the dependency pattern, and I have not verified it verbatim against the granted claim set in USPTO PatentCenter. Treat the exact language of claims 1–5 as needing confirmation.

5. Litigation / post-grant activity

• PTAB IPR2021-00330 — Toshiba America Electronic Components, Inc. et al. v. Monument Peak Ventures, LLC, filed 2020-12-21, on U.S. Pat. No. 7,583,294, Tech Center 2600. Status is reported inconsistently across sources: Google Patents says "Pending – Instituted," while the IPVerse/Greyb PTAB tracker lists "Institution Denied" with an institution decision date of 2021-06-25. I cannot resolve this conflict from available sources; verify at the PTAB docket.
• District court — Monument Peak Ventures suit in the California Central District Court, case 8:19-cv-02181 (per Google Patents litigation data).
• Federal Circuit — Google Patents lists CAFC case 21-1052 (a 2021 appeal, consistent with the 2019–2021 district court and IPR activity). I found no 2026 CAFC docket entry, briefing, or decision for US 7,583,294 in any search performed.
• The patent is expired (adjusted expiration 2022-01-06), which makes fresh 2026 appellate activity unlikely absent a damages/back-damages tail from earlier litigation.

Bottom line on the 2026 search request: based on the searches I was able to run, there is no evidence of a 2026 CAFC docket involving US 7,583,294; the only appellate identifier surface is the 2021 appeal (21-1052), and the only PTAB proceeding is IPR2021-00330.

I'll research litigation involving US patent 7583294 specifically.

Let me verify the outcomes of the IPR and the Federal Circuit appeal, and check for any additional assertions of this patent.

Litigation Involving U.S. Patent No. 7,583,294

Patent identified: U.S. Patent No. 7,583,294 (US7583294B2), "Face Detecting Camera and Method," invented by Lawrence A. Ray and Henry Nicponski; original assignee Eastman Kodak Co.; currently held by Monument Peak Ventures, LLC. Application No. 11/122,593; priority date Feb. 28, 2000; granted Sept. 1, 2009; expired/expiration adjusted to Jan. 6, 2022.

I found one district-court action (plus its Federal Circuit appeal) and one PTAB inter partes review specifically naming U.S. Patent No. 7,583,294. Details below, followed by sources and caveats.

1. District Court — Monument Peak Ventures, LLC v. Toshiba

Outcome detail: The complaint asserted several Kodak-derived patents, including U.S. Patent Nos. 6,903,762; 7,177,484; 7,583,294; 7,684,090; 8,964,064; and 9,549,095. On August 12, 2020, Judge Carter granted the defendants' motion to dismiss (Doc. 44), holding the asserted claims (including those of the '294 patent) patent-ineligible under 35 U.S.C. § 101 under the Alice two-part framework. A Stipulated Final Judgment (Doc. 51) was entered September 11, 2020, dismissing the '090 claims with prejudice and entering judgment in favor of all Toshiba defendants against Monument Peak Ventures on the § 101 ineligibility of the adjudicated claims of the '294 patent (among others). Each party bore its own fees and costs. The case ended at the pleading stage — there was no trial, verdict, injunction, or damages award.

2. Federal Circuit Appeal — Monument Peak Ventures, LLC v. Toshiba

The appeal challenged (1) the August 12, 2020 Order Granting Defendants' Motion to Dismiss (Doc. 44) and (2) the September 11, 2020 Final Judgment (Doc. 51).

⚠️ I could not confirm the ultimate Federal Circuit disposition from the sources retrieved. The most recent status I could verify is "Pending." You should check the Federal Circuit docket for 21-1052 directly to confirm whether it was affirmed, dismissed, or otherwise resolved. I do not want to state an outcome I could not verify.

3. PTAB Inter Partes Review — Toshiba v. Monument Peak Ventures

⚠️ Conflict flag (per operating rules): Two sources disagree on the PTAB outcome. The GreyB/IPVerse PTAB database entry states the institution decision was issued June 25, 2021 with status "Institution Denied." The Google Patents page's litigation summary labels the proceeding "Pending – Instituted." I was unable to run a further search to resolve this conflict, so I present both rather than choosing. Given the final judgment in the co-pending district court case (Sept. 2020) and the § 314(a) discretionary-denial arguments commonly raised, "Institution Denied" is a plausible outcome, but please verify against PTAB's official record (IPR2021-00330).

Sources

• Google Patents, US7583294B2 (litigation links): https://patents.google.com/patent/[US7583294](/patent/US7583294)/en
• Unified Patents patent page: https://portal.unifiedpatents.com/patents/patent/US-7583294-B2
• RPI Empower (relevant context for MPV's assertion campaign): https://litigation.rpxcorp.com/news/81877-earlier-2024-case-bogged-down-over-service-of-process-monument-peak-ventures-sues-vivotek
• PTAB documents via Docket Alarm (IPR2021-00330): https://www.docketalarm.com/search/PTAB/ (case IPR2021-00330)
• GreyB/IPVerse IPR entry: https://ipverse.greyb.com/ptab-web/cases/case-details/IPR2021-00330
• District-court filings (Notice of Appeal, Stipulated Final Judgment, Order on Motion to Dismiss) reproduced at PTACTS: https://ptacts.uspto.gov/ptacts/public-informations/petitions/[1544742](/patent/1544742)/

Important scoping notes (per your strict rules)

1. Do not confuse with the '328 patent. One search result referenced Facet Tech. Corp. v. Mobileye Glob., Inc., No. 2:24-cv-00058 (E.D. Tex.), concerning U.S. Patent No. 7,328,328 ("the '328 Patent"). That is a different patent number and is excluded from this answer.

2. Do not confuse with other Monument Peak Ventures suits. Actions such as Monument Peak Ventures v. Open Text (2:25-cv-00521, E.D. Tex.) and Monument Peak Ventures v. Vivotek (6:24-cv-00361, W.D. Tex.) involve other patents (e.g., 8,024,311; 7,035,461; 8,665,345; 8,643,746; 9,013,604) — not U.S. 7,583,294. They are excluded.

3. The Japanese document JP7583294B2 (a refractory-brick composition patent, Shinagawa Refractories) is an unrelated foreign document that merely happens to share the numeral "7583294." It is not related to U.S. 7,583,294.

4. No record found of any litigation involving U.S. 7,583,294 other than the Toshiba district-court action (8:19-cv-02181) and its Federal Circuit appeal (21-1052), plus the Toshiba PTAB petition (IPR2021-00330). All three arose from a single dispute between Monument Peak Ventures and Toshiba.

5. Unverified items: the final Federal Circuit disposition (21-1052) and the exact PTAB institution outcome (see conflict flag above). I recommend confirming both on PACER/CM-ECF and the Federal Circuit/PTAB dockets, respectively.

I'll start by pulling up the PTAB and litigation references surfaced on the patent page, then verify each against public sources.

Found a hit: IPR2021-00330 (Toshiba v. Monument Peak Ventures). The ODP block says "no proceedings," so I need to reconcile that. Let me dig into the institution decision, POP request, and the CAFC appeal.

Excellent — the picture is emerging. Let me nail down the petition grounds/art and confirm whether any other IPR, CBM, or PGR exists on this patent.

Proceedings overview

One AIA trial proceeding exists on US 7,583,294 — IPR2021-00330 (Toshiba v. Monument Peak Ventures) — and it never reached trial: institution was denied (twice, on different grounds), so zero claims were canceled and zero were sustained by the PTAB, and no Final Written Decision exists. The patent's actual death blow came from the parallel district court, which held the '294 patent invalid under 35 U.S.C. § 101, a judgment affirmed by the Federal Circuit on 2021-12-13 (Appeal No. 2021-1052) — that judgment is what mooted Toshiba's IPR. For a defendant today: there is no PTAB invalidation to stand on and no § 315(e) estoppel against you, but the patent expired 2022-01-06 and its asserted claims were adjudicated invalid under § 101. This is a patent you defeat with the district-court § 101 record, not with the PTAB record.

⚠️ Flag on the ODP data: The structured "PTAB proceedings on file" block reports no AIA proceedings. That is inconsistent with the public record. IPR2021-00330 is documented on the face of the Google Patents page for this patent (PTAB case link), on the PTAB docket aggregators, in the POP's own precedential decision, and in the petition exhibits. It appears the ODP ingest missed this case (a 2020 petition that terminated in 2022). Do not rely on "no PTAB activity" for this patent. Also note Google Patents labels IPR2021-00330 "Pending - Instituted" — that flag is stale and substantively wrong; the case was never instituted and is long terminated.

IPR2021-00330 — Toshiba America Electronic Components, Inc. v. Monument Peak Ventures, LLC

• Type: Inter Partes Review
• Filed: 2020-12-21 (as docketed; petitioner moved to correct the filing date to 2020-12-16, the Fedwire payment date — see below)
• Status: Institution Denied (third-party structured record: Docket Alarm and GreyB/IPVerse both report "Institution Denied"; Docket Alarm lists termination 2021-06-25). Verbatim caveat: the ODP structured block in this prompt lists no proceeding at all.
• Judge panel: Motion/institution panel — Joni Chang, Juliet Mitchell Dirba, Michael Zecher (per Docket Alarm PTAB docket). POP panel (2022-01-14 decision) — Scott R. Boalick (Chief APJ, author), Andrew Hirshfeld (Director, performing the functions of the Under Secretary/Director), Jacqueline Wright Bonilla (Deputy Chief APJ), with Hirshfeld and Bonilla concurring.
• Petition grounds: Challenged claims 1, 5, 10, 11, and 19 (recited verbatim in the POP decision). Grounds were § 103 obviousness over prior-art patents/printed publications. Petitioner's own framing: the challenged claims "relate to detecting faces in an image by utilizing two algorithms, in which the second algorithm incorrectly detects faces—i.e., generates 'false positives'—at a lower rate than the first algorithm … the prior art analyzed below establishes that such algorithmic methods for face detection were well-known long before the '294 patent was filed." I could not verify the specific reference names (Ex. 1005, 1006, etc.) from the sources retrieved, so I am not naming them rather than guessing.
• Institution decision: 2021-06-25 — DENIED. A split panel held Toshiba had not established that the required fee accompanied the petition before the § 315(b) one-year deadline (complaint served on TAEC 2019-12-17; deadline 2020-12-17; docketed filing 2020-12-21). The majority rejected Toshiba's showing based on a Fedwire wire-transfer confirmation and denied Toshiba's motion to correct the filing date. The dissent would have found the regulatory requirements satisfied, or good cause to excuse the rules under 37 C.F.R. §§ 42.5(b), 42.5(c)(3).
• Final Written Decision: None. No trial was ever instituted, so no claims were canceled, no claims were held patentable, and no claim-level merits ruling exists. Any statement that the PTAB canceled or sustained claims of the '294 patent would be false.
• Settlement / termination: No settlement — the parties litigated the filing-date issue aggressively (patent owner's preliminary response argued both § 315(b) time bar and Fintiv/§ 314(a) discretionary denial; petitioner sought rehearing and POP review).
• Post-denial procedural history (this is the important part):
  • 2021-07-23 — Petitioner's motion for rehearing under 37 C.F.R. § 42.71(d) (Paper 10).
  • 2021-07-28 — Board notice of receipt of a Precedential Opinion Panel (POP) request (Paper 11; Ex. 3003).
  • 2021-09-24 — POP granted review (Paper 13).
  • Oct–Nov 2021 — POP briefing: PO Opening Brief (Paper 15, 2021-10-22); Petitioner's Response (Paper 14); PO Reply (Paper 17, 2021-11-05). The patent owner argued a Fedwire confirmation is not evidence of receipt of payment and that wire transfers are made at the payor's risk.
  • 2022-01-14 — POP decision (Paper 19), designated precedential: The POP granted rehearing, held that a petitioner's Fedwire payment confirmation "constitute[s] sufficient evidence that the required fee accompanied the petition" and is sufficient evidence that payment was "received" under 37 C.F.R. § 42.103(b), relying on UCC Article 4A § 4A-406 (the originator pays the beneficiary when the beneficiary's bank accepts the payment order) and rejecting the earlier majority's reading. Toshiba's filing date was therefore corrected to 2020-12-16, and the panel vacated the § 315(b) time-bar denial. However, the POP nevertheless declined to institute, citing intervening events — the district court's § 101 invalidity judgment, which the Federal Circuit had affirmed on 2021-12-13. (Public summaries describe this as a discretionary denial on the changed circumstances, not a merits ruling.)
  • 2022-01-25 — Board notice approving refund of the filing fee (Paper 22), consistent with institution never having occurred.
• Appeal: No appeal of the IPR — an institution denial is unreviewable under § 314(d) (Cuozzo, Thryv), and none was taken. Separately, the district court judgment was appealed: Monument Peak Ventures, LLC v. Toshiba America Business Solutions, Inc., Toshiba America Electronic Components, Inc., Toshiba Corporation, No. 2021-1052 (Fed. Cir.), filed 2020-10-15, decided 2021-12-13, per curiam (Lourie, Clevenger, Prost) — AFFIRMED under Fed. Cir. R. 36 (nonprecedential summary affirmance, no opinion). Underlying: C.D. Cal. No. 8:19-cv-02181-DOC-DFM (Judge David O. Carter); § 101 invalidity order 2020-08-12, stipulated final judgment 2020-09-11.
• Defensive value: The PTAB gives you nothing offensively — there is no IPR estoppel under § 315(e)(2) (estoppel attaches only after a final written decision), and no PTAB claim cancelation to cite. What you do have is a Federal Circuit-affirmed judgment that the '294 patent is invalid under § 101, plus a patent that expired 2022-01-06. Also useful: Toshiba is now itself time-barred under § 315(b), and the eve-of-deadline filing that created this mess is a cautionary case study in how a one-day deadline miss nearly became a one-year fatal bar.

Strategic summary

Claim status: CANCELED — none (PTAB never instituted); SUSTAINED — none (no FWD ever issued); UNTESTED at the PTAB — claims 1, 5, 10, 11, and 19 were the only claims challenged, and no claim reached a merits decision. On the court side, the C.D. Cal. entered judgment that the '294 patent is invalid under § 101, affirmed on 2021-12-13. Note carefully the limits of that: a district court invalidity holding is inter partes — it binds Monument Peak Ventures and those parties, and carries strong stare-decisis/issue-preclusion weight, but it is not an in rem cancelation of the claims. There is also no certificate of correction, disclaimer, or reexamination certificate canceling claims. So the correct defensive characterization is "invalidated by a final, affirmed district court judgment," not "claims canceled by the USPTO."

Estoppel landscape: A defendant today is in an unusual and favorable posture — § 315(e)(2) estoppel never attached to anyone, because IPR2021-00330 was never instituted and no final written decision issued. Any defendant (including non-Toshiba parties) remains free to file its own IPR on any art, since Toshiba's petition produced no estoppel and no § 325(d) record admissions binding on others. Conversely, the flip side of no-institution is that there is no PTAB invalidation to borrow. Separately, Toshiba and its real parties in interest/privies are now time-barred under § 315(b) (served 2019-12-17; the year ran out 2020-12-17). One practical nuance: parties previously sued by MPV on this patent and served more than a year ago are also time-barred from filing an IPR — for those defendants, the § 101 judgment and the expiration date are the whole game.

Pattern signals: The same petitioner has not filed multiple IPRs on this patent — one petition, one proceeding. MPV litigated hard but its energies were spent in district court (the POP brief records the C.D. Cal. proceedings, including the § 101 win for Toshiba). No defensive aggregator (e.g., Unified Patents) appears as a petitioner — the Google Patents litigation link to "Unified Patents PTAB Data" is a data source, not an indication that Unified filed; the real petitioner was Toshiba, a true-party defendant. On the monetization side, the Unified Patents patent page lists MPV's parent company as Dominion Harbor Enterprises LLC, i.e., this is a Dominion Harbor-affiliated NPE monetization posture on a legacy Kodak portfolio — expect assertion/late-life campaign behavior rather than manufacturing practice. The patent is now expired (2022-01-06), which caps any recovery to past damages and eliminates injunctive leverage.

Recommended next steps

1. Lead with the § 101 judgment, not the PTAB. The POP's own reasoning in IPR2021-00330, Paper 19 (2022-01-14) relied on the intervening § 101 invalidity. Obtain the C.D. Cal. final judgment (8:19-cv-02181-DOC-DFM, entered 2020-09-11) and the Federal Circuit's 2021-12-13 Rule 36 judgment in No. 2021-1052, and quote the affirmance verbatim: "PER CURIAM (LOURIE, CLEVENGER, and PROST, Circuit Judges). AFFIRMED. See Fed. Cir. R. 36." That is your dispositive invalidity citation against this patent.
2. Do not cite the PTAB for anything except the filing-date precedent. IPR2021-00330's precedential contribution is about Fedwire payments and 37 C.F.R. § 42.103(b) — not about the patentability of face-detection claims. If opposing counsel claims "the PTAB upheld these claims," that is false; there was no institution and no FWD.
3. Assert the § 315(b) bar for Toshiba and any prior-served privies, and confirm the expired status (2022-01-06) to eliminate forward-looking relief. Confirm current assignee status — Google Patents lists Monument Peak Ventures LLC, with a 2023-08-15 release by secured party involving Intellectual Ventures Fund 83 LLC; verify the chain before responding to any demand.
4. If your client is a new defendant (never served by MPV on this patent and not a Toshiba privy), you retain a full, un-estopped IPR option — but weigh it against the fact that the patent is expired and already carries an affirmed § 101 invalidity judgment. On these facts, an IPR is likely an unnecessary expense; a motion practice built on § 101 preclusion plus expiration is the cheaper path.
5. Verify against primary sources before filing. Pull the full IPR2021-00330 file (Papers 9, 10, 11, 13, 15, 17, 18, 19, 22) from PTAB E2E, and confirm the exact grounds/reference names in the petition, which I could not independently verify here.

Sources

• Google Patents, US 7,583,294 B2 — https://patents.google.com/patent/US7583294/en (PTAB case link, litigation entries, expiration 2022-01-06)
• PTAB docket, IPR2021-00330 — https://www.docketalarm.com/cases/PTAB/IPR2021-00330/Toshiba_America_Electronic_Components_Inc._v._MONUMENT_PEAK_VENTURES_LLC/ (papers 8–22, judges, "Institution Denied")
• POP decision coverage, Toshiba America Electronic Components, Inc. v. Monument Peak Ventures, LLC, IPR2021-00330, Paper 19 (PTAB Jan. 14, 2022) — Element IP summary: https://www.elementiplaw.com/ptab-precedential-opinion-holds-fedwire-payment-confirmation-sufficient-evidence-of-payment/ ; public copy of the decision: https://www.jetro.go.jp/ext_images/world/n_america/us/ip/pdf/Japanese_translation/20220114.pdf
• Patent owner's POP briefs and preliminary response, IPR2021-00330 — https://www.docketalarm.com/cases/PTAB/IPR2021-00330/Toshiba_America_Electronic_Components_Inc._v._MONUMENT_PEAK_VENTURES_LLC/docs/04-02-2021-Patent_Owner/Preliminary_Response-8-PO_Preliminary_Response.pdf
• CAFC No. 2021-1052, judgment of 2021-12-13 — https://cases.justia.com/federal/appellate-courts/cafc/21-1052/21-1052-2021-12-13.pdf ; docket: https://dockets.justia.com/docket/circuit-courts/cafc/21-1052

Uncertainties I am flagging rather than papering over: (a) the ODP block says no proceedings — contradicted by the public record; (b) the exact prior-art references in Toshiba's petition grounds could not be verified; (c) whether the POP's final refusal to institute was formally styled a § 314(a) discretionary denial or an order declining to institute after vacatur — public summaries describe it as a denial driven by the intervening affirmed § 101 judgment, and the 2022-01-25 fee refund confirms no trial occurred; (d) the specific claim numbers the district court held invalid under § 101 were not verifiable from the retrieved sources.

I'll research the assignment chain for US 7,583,294, starting with the USPTO assignment records and cross-referencing the legal events already surfaced.

Let me try to pin down the actual reel/frame numbers and correspondents.

Let me try to nail down the reel/frame numbers and correspondents.

Assignment Record Reconstruction — US 7,583,294 B2

⚠️ Data-availability caveat, stated up front (per operating rules): I could not complete a live query of USPTO Assignment Center (assignmentcenter.uspto.gov) in this session, and the web searches available to me did not return the reel/frame numbers or correspondent-of-record fields for this patent's assignment chain. Google Patents' "Reassignments" panel mirrors the USPTO assignment data but the fetched page text does not expose reel/frame or correspondent. Accordingly:

• The dates, conveyance types, assignors and assignees below are taken from the Google Patents legal-events/assignment mirror (which re-publishes USPTO assignment records) and corroborated by the SEC 8-K, Kodak's 2013 annual report, and RPX/Lexology reporting.
• Reel/frame numbers and correspondent attorneys are NOT fabricated here. They are marked "not retrieved." This is the single most important gap in this analysis — the correspondent field is exactly the tell the task wants, and I could not get it. Verify at https://assignmentcenter.uspto.gov/ by searching patent number 7583294.

The chain is a real, documented, four-link ownership transfer with one security-interest encumbrance and one secured-party release — not a case of "no records."

Inventors

Unusual pattern — flag: The inventors' assignment to Eastman Kodak was only recorded 2013-08-07 (Google Patents legal events), i.e. ~13 years after the Feb. 28, 2000 priority filing, and — critically — ~3.5 months after the patent had already been assigned by Kodak to Intellectual Ventures Fund 83 (recorded 2013-04-24). This ordering is a chain-of-title cleanup: the inventor→Kodak link was papered/recorded contemporaneously with the bankruptcy portfolio transfer to perfect the record for the buyer. It is not evidence of inventors leaving Kodak.

No evidence found of either inventor departing Kodak within 12 months of filing. Do not infer the "all-inventors-depart" fire-sale precursor from this record — the 13-year recordation gap is a documentary artifact, not a personnel signal.

Original assignee

Eastman Kodak Company (original assignee per Google Patents; Rochester, New York).

• Primary line of business: Integrated imaging — photographic film, photofinishing, and (relevant here) digital still cameras and image-processing firmware/software. The '294 disclosure is squarely a Kodak digital-camera feature set (face detect → AF/AE/flash/composition).
• Product embodiment: Kodak did market consumer digital cameras in the 2000s (EasyShare line) that performed face- and skin-tone-based exposure/color work; however, I have no specific evidence tying a named shipping product to the literal claims of the '294 patent. Treat "shipped a product embodying the claims" as not established for this record.
• Current status: Chapter 11 bankruptcy — Kodak filed in the Southern District of New York in January 2012, sold the digital-imaging portfolio in the §363 sale (Feb. 2013), and emerged September 2013 as a commercial-printing/advanced-materials company. Kodak (NYSE: KODK) survives as an operating entity but exited the consumer digital camera business in 2012. The '294 patent did not stay with it.

Assignment timeline

(Reel/frame and correspondent columns could not be retrieved — see caveat above. Conveyance types are as titled in the Google Patents assignment mirror.)

• 2000-02-28 (executed, original) — Reel not retrieved

  • Conveyance: Assignment (inventors → Eastman Kodak) — but see below; recorded late
  • Assignor: Lawrence A. Ray; Henry Nicponski
  • Assignee: Eastman Kodak Company
  • Correspondent: not retrieved
  • Context: Original employer assignment of the priority application (Ser. No. 09/514,436).

• 2012-02-21 (recorded) — Reel not retrieved

  • Conveyance: Security Interest ("SECURITY INTEREST (SEE DOCUMENT FOR DETAILS)")
  • Assignor: Eastman Kodak Company; Pakon, Inc.
  • Assignee: Citicorp North America, Inc., as Agent
  • Correspondent: not retrieved
  • Context: Securitization / DIP-lender collateral — Kodak's Chapter 11 secured-credit facility taking a lien over the IP estate. Not a change of ownership.

• 2013-02-01 (recorded) — Reel not retrieved

  • Conveyance: Release (Patent Release / secured-party lien release)
  • Assignor: Citicorp North America, Inc.; Wilmington Trust, National Association
  • Assignee: Kodak Portuguesa Limited; Eastman Kodak International Capital Company, Inc.; Kodak Americas, Ltd.; Far East Development Ltd.; Eastman Kodak Company; Kodak Aviation Leasing LLC; Kodak (Near East), Inc.; Kodak Realty, Inc.; Kodak Imaging Network, Inc.; Creo Manufacturing America LLC; FPC Inc.; Kodak Philippines, Ltd.; Pakon, Inc.; NPEC Inc.; Laser-Pacific Media Corporation; Qualex Inc.
  • Correspondent: not retrieved
  • Context: Lien discharge ahead of the free-and-clear §363 sale; coincides with the Feb. 1, 2013 closing of the Kodak/IV digital-imaging transaction (Kodak 2013 annual report).

• 2013-04-24 (recorded) — Reel not retrieved

  • Conveyance: Assignment of Assignors' Interest
  • Assignor: Eastman Kodak Company
  • Assignee: Intellectual Ventures Fund 83 LLC
  • Correspondent: not retrieved
  • Context: Bankruptcy fire-sale / transfer to aggregator. Kodak's digital-imaging portfolio (~1,100 assets) sold to IVF83 for ~$527M, court-approved under §363; IVF83 was funded by an IV/RPX-organized licensee consortium (Apple, Google, Samsung, Microsoft, Amazon, Adobe, Facebook, FUJIFILM, Huawei, HTC/RIM, Shutterfly). See SEC Form 8-K filed 2012-12-21.

• 2013-08-07 (recorded) — Reel not retrieved

  • Conveyance: Assignment of Assignors' Interest
  • Assignor: Nicponski, Henry; Ray, Lawrence A. (the inventors)
  • Assignee: Eastman Kodak Company
  • Correspondent: not retrieved
  • Context: Chain-of-title cleanup — 13-year-late recordation of the original inventor→Kodak assignment, filed after the sale to IVF83. ⚠️ Contradiction/ordering flag against the prior section: the earlier bibliographic table listed this 2013-08-07 event without noting that it postdates the 2013-04-24 Kodak→IVF83 transfer. That ordering matters and should be surfaced, not smoothed over.

• 2017-03-09 (recorded; executed ~Feb. 2017) — Reel not retrieved

  • Conveyance: Assignment of Assignors' Interest
  • Assignor: Intellectual Ventures Fund 83 LLC
  • Assignee: Monument Peak Ventures, LLC
  • Correspondent: not retrieved
  • Context: Transfer to asserter. Dominion Harbor Enterprises (DHE) announced 2017-02-16 the acquisition of 1,000+ former-Kodak families from IV, to be commercialized by DHE subsidiary Monument Peak Ventures. MPV was incorporated in Texas in December 2016 by Dominion's litigating affiliate Monument Patent Holdings, LLC (RPX/Lexology reporting).

• 2023-08-15 (recorded) — Reel not retrieved

  • Conveyance: Release by Secured Party
  • Assignor: Intellectual Ventures Fund 83 LLC
  • Assignee: Monument Peak Ventures, LLC
  • Correspondent: not retrieved
  • Context: Encumbrance release, not a transfer. Consistent with discharge of a seller-financing/security interest running from the IV side over MPV-held assets. Ownership remains with MPV; this is the only post-2017 event.

Terminal owner: Monument Peak Ventures, LLC (current assignee per Google Patents; parent entity reported as Dominion Harbor Enterprises LLC per Unified Patents). Patent expired — adjusted expiration 2022-01-06 (Google Patents) / 2022-01-05 (Unified Patents).

Timeline diagram

timeline
    title Ownership of US 7583294
    2000 : Filed by Eastman Kodak
    2009 : Patent issued
    2012 : Citicorp security interest recorded
         : Kodak files Chapter 11
    2013 : Liens released
         : Assigned to Intellectual Ventures Fund 83
         : Inventor assignment to Kodak recorded
    2017 : Assigned to Monument Peak Ventures
    2019 : First suit on this patent
    2022 : Patent expired
    2023 : Security interest release to MPV

NPE / troll-pattern signals

1. Shell-entity transfer — PRESENT. Two consecutive operating-to-licensing hops: Kodak → Intellectual Ventures Fund 83 LLC (rec. 2013-04-24) and IVF83 → Monument Peak Ventures, LLC (rec. 2017-03-09). MPV is a Texas single-purpose LLC incorporated December 2016 by Monument Patent Holdings, LLC, with no products in commerce; it began its first suits in March 2018. Corroborated by RPX/Lexology (MPV is a Dominion Harbor monetization vehicle).

2. Known asserter in the chain — PRESENT (twice over). (a) Intellectual Ventures is on the enumerated NPE list and held the patent 2013–2017. (b) Monument Peak Ventures / Dominion Harbor Enterprises is a high-frequency plaintiff tracked by RPX and Unified Patents; Dominion's founder/principal David Pridham is the co-founder and former CEO of IP Navigation Group (IPNav) — also on the enumerated list. Both signal and named-list match are satisfied.

3. Repeat correspondent across the chain — UNCLEAR / unverified. I could not retrieve the correspondent-of-record for any link, so I cannot test for a recurring recording attorney. This signal cannot be scored. It is the highest-value unresolved item; the Kodak→IVF83 and IVF83→MPV recordings are the two links most likely to share a single monetization-side filer.

4. Cascading transfers (<24 months through chained LLCs) — NOT PRESENT (as defined), but note a same-year cluster. The LLC-to-LLC hop (IVF83→MPV) is a single step ~4 years after the Kodak→IVF83 transfer — outside the <24-month framing. There is a 2013 cluster (release 2013-02-01; Kodak→IVF83 2013-04-24; inventor→Kodak 2013-08-07) inside ~6 months, but those are one sale plus its title cleanup, not serial LLC layering.

5. Pre-litigation transfer within 6 months of suit — PRESENT, but only against the portfolio's assertion date, not this patent's. MPV acquired IVF83's Kodak assets ~Feb. 2017 and filed its first MPV lawsuits in March 2018 (~13 months later). For this patent specifically, the first suit is the Nov. 2019 Toshiba action (see the prior Litigation section, 8:19-cv-02181) — ~33 months after acquisition, outside 6 months. Score the "transfer arranged to enable this suit" reading as not present; the acquisition was plainly a portfolio-wide assert-ready purchase.

6. Bankruptcy fire-sale — PRESENT (textbook). Originator Eastman Kodak filed Chapter 11, and the patent moved out via the §363 sale to IVF83 for ~$527M, approved by the Bankruptcy Court (S.D.N.Y.) with closing Feb. 1, 2013 and the assignment recorded 2013-04-24. Corroborated by Kodak SEC Form 8-K (2012-12-21) and Kodak's 2013 annual report ("On February 1, 2013, Kodak entered into a series of agreements related to the monetization of certain of its intellectual property assets…").

7. Privateering — NOT PRESENT. Kodak was a distressed seller exiting the business, not an operating company arming an NPE to attack competitors. The IV/RPX consortium's participation (competitor licensees funding the buy) is closer to a defensive/royalty-free clearing move at the 2013 stage; the aggressor posture only appears at the 2017 IV→MPV step.

8. Defensive aggregator at the end of the chain — NOT PRESENT. Although IV and RPX were involved in the 2013 acquisition (and RPX is a defensive-aggregation player), the chain does not terminate at RPX/AST/LOT/Unified/OIN. It terminates at Monument Peak Ventures, an asserting entity, which then sued Toshiba in 2019 and was hit with IPR2021-00330. The defensive layer was bypassed three years later when IV resold to Dominion.

Verdict

NPE — high confidence.

Two or more strong signals are met on the record: (i) a documented operating-to-licensing shell transfer chain — Kodak → Intellectual Ventures Fund 83 LLC (assignment recorded 2013-04-24) → Monument Peak Ventures, LLC (recorded 2017-03-09), the latter a Texas single-purpose LLC with no products; (ii) a named-list asserter on both ends (Intellectual Ventures, and Dominion Harbor / Monument Peak Ventures, whose principal is IPNav's co-founder); and (iii) a textbook bankruptcy fire-sale origin (Kodak Chapter 11, §363 sale to IVF83, ~$527M). The chain then manifests as concrete assertion activity — the Nov. 2019 Toshiba district-court suit (C.D. Cal. 8:19-cv-02181) and IPR2021-00330 (both covered in the prior Litigation section) — with the patent now expired (adj. exp. 2022-01-06).

One caveat on the verdict's completeness: because the correspondent-of-record and reel/frame fields were not retrievable, signal #3 (repeat correspondent) is unscored, and every link above is dated by its USPTO recording date, which for the 2013 links trails execution. Re-run against the Assignment Center to capture reel/frame and correspondents — that is the evidence that would let this be upgraded from "high confidence" to "fully documented."

Verification link: https://assignmentcenter.uspto.gov/ (search patent number 7583294; the same records underlie https://assignment.uspto.gov/patent/index.html).

I'll research the prior art cited for US 7,583,294, starting with the patent's own front-page citations.

I found a partial citation list. Let me get the complete list of cited references for US 7,583,294.

Let me get the remaining citations in the list.

Scope and verification note (read first)

• Target document: U.S. Patent No. 7,583,294 B2 ("Face detecting camera and method"). It is a continuation of Ser. No. 09/514,436 (issued as U.S. Pat. No. 6,940,545), so the two share an identical specification, priority date (2000‑02‑28), and essentially the same "References Cited" front page. Where I cite a front‑page reference, it applies to both.
• The "References Cited" list I could retrieve is incomplete. Google Patents reports "Patent Citations (32)" for this family, and Unified Patents reports "Patent Art (33)" + "Non‑Patent Literature (3)". I was able to retrieve and verify 17 of the 32 U.S./foreign patent citations. I could not retrieve the remaining ~15 in this session. I list the 17 I verified below and explicitly flag the gap rather than guess at the rest.
• Important legal caveat on the framing of this task: These are references the Examiner cited on the face of the patent. The patent nevertheless issued on 2009‑09‑01, which means the Examiner did not treat any of them as anticipatory under § 102. They are therefore properly understood as § 103 (obviousness) art and background art, not as § 102 anticipatory references. Stating that a cited reference "anticipates" a claim would be inconsistent with the fact that the patent granted over the entire set. I have therefore framed the mapping as "which claims this reference is relevant to / could be combined against," and I state plainly where a single reference cannot anticipate for want of an element.

1. The three independent claims these references would be measured against

From the previously generated analysis (authoritative here), the independent claims are:

• Claim 1 — Method: (a) prescreen image data with a first algorithm that determines face candidates via a pattern‑matching technique identifying image windows likely to contain faces based on color and shape; (b) operate on those candidates with a second algorithm using a posterior‑probability‑function classifier.
• Claim 5 — Method claiming the same two‑stage architecture plus the express limitation that the first algorithm has a first (higher) false‑positive rate and the second has a lower false‑positive rate.
• Claim 19 — Computer program product on a computer‑readable storage medium performing steps (a)+(b) plus the first‑higher/second‑lower false‑positive‑rate limitation.

Dependent structure: claims 2–4 depend from claim 1 (inferred); claims 6–18 depend from claim 5 (including apparatus‑flavored dependents 6, 13, 14 covering camera/face‑data/storage, orientation (13), and red‑eye + display warning (14); 16–17 add red‑eye corroboration by face presence).

Critical § 102 point: For any single reference to anticipate claim 1/5/19, it must disclose both a color‑and‑shape prescreener and a posterior‑probability classifier. None of the cited art does both in one document. That is precisely why the combination survived examination.

2. Patent citations for US 7,583,294 (verified 17 of 32)

Gap flag: ~15 further patent citations (positions 18–32 on the Google Patents "Patent Citations (32)" list) were not retrievable in this session and are not represented above. I will not speculate about their contents. If completeness matters for a validity opinion, the remaining entries must be pulled directly from the '294 front page in USPTO PatentCenter or Google Patents.

3. Non‑patent literature cited (the 3 NPL items, plus textbook)

These are the most technically relevant references in the entire set, because the patent names them as the sources of its two component algorithms:

4. Assessment — most relevant prior art and § 102 bottom line

Ranking of relevance to the actual claims:

1. Wu et al. (1999) — the closest art to claim 1(a) and the "first algorithm" of claim 5.
2. Schneiderman & Kanade (1998) — the closest art to claim 1(b) and claim 19's classifier.
3. US 5,835,616 ("Face detection using templates") — closest patent to claim 1(a).
4. US 5,710,833 ("probabilistic eigenspace analysis") — closest patent to claim 1(b)/19's posterior‑probability classifier.
5. US 5,432,863 (Kodak red‑eye) and US 5,629,752 (Fuji face‑based exposure) — closest to the camera‑function dependents (claims 14, 16, 17, 6, 13).

§ 102 conclusion: On the retrieved record, no single cited reference, taken alone, anticipates any of independent claims 1, 5, or 19. Each of the two most technically relevant references (Wu; Schneiderman & Kanade) discloses only one of the two claim‑1 stages, and the novelty of the asserted claims resides specifically in the two‑stage combination plus the deliberate false‑positive‑rate asymmetry recited in claims 5 and 19 (the "almost‑never‑miss" prescreener feeding a "very‑low‑false‑positive" confirmer). The cited art is therefore § 103 combination material, not anticipatory art — consistent with the fact that the patent issued over all of it.

One explicit contradiction to flag with the earlier sections: the earlier section reported the PTAB IPR2021‑00330 on this patent with a conflicting institution status ("Institution Denied" per GreyB/IPVerse vs. "Pending – Instituted" per Google Patents) and an unresolved CAFC outcome for 21‑1052. That conflict remains unresolved here; nothing in this prior‑art search resolves it, and it does not change the § 102 analysis above (the district court disposed of the case on § 101 eligibility, not on § 102/§ 103 prior art).

Sources

• Google Patents, US7583294B2 (patent citations, NPL, family): https://patents.google.com/patent/US7583294/en
• Google Patents, US6940545B1 family page ("Patent Citations (32)"): https://patents.google.com/patent/[US6940545](/patent/US6940545)
• Unified Patents page (Patent Art 33 / NPL 3 / Referenced By 45): https://portal.unifiedpatents.com/patents/patent/US-7583294-B2
• Full patent text you supplied (References/non‑patent citations of Wu et al.; Schneiderman & Kanade; Grill & Scanlon; Lim; Visionics Face‑It).
• IPVerse/GreyB IPR2021‑00330: https://ipverse.greyb.com/ptab-web/cases/case-details/IPR2021-00330

Uncertainty flags (per operating rules): (i) I verified 17 of 32 patent citations; the remaining ~15 are unretrieved and unstated; (ii) the exact wording of claims 1–5 was inferred in the earlier section and was not re‑verified against PatentCenter here; (iii) item 17's title is partially truncated in the source snippet ("…detection of…"); (iv) no 2026 docket activity for US 7,583,294 was found, consistent with the earlier section.

I'll ground this analysis by pulling the prior-art references cited in the patent and checking what prosecution/prior-art record exists for this family.

Let me pull the granted claim set for the '294 continuation specifically, since the prior section flagged claims 1–5 as inferred.

Cross-reference flags before the analysis

Contradiction flagged (important). The earlier "Patent summary" section stated that the exact wording of claims 1–5 of US 7,583,294 was "inferred" and that the false-positive-rate limitation appeared in claim 5 (characterizing claim 1 as the plain two-step method). I have now verified the granted text of claim 1 directly, and the earlier inference is partly wrong. Per RPX Insight's reproduction of the granted claim set (https://insight.rpxcorp.com/patent/[US7583294B2](/patent/US7583294B2)), claim 1 does recite the false-positive-rate relationship, and it also recites a structural element ("by using an algorithm interface adapter") that the summary did not mention:

Claim 1. "A method for determining the presence of a face from image data, said method comprising the steps of: (a) prescreening the image data with a first algorithm by using an algorithm interface adapter, the first algorithm determining a plurality of face candidates utilizing a pattern matching technique that identifies image windows likely to contain faces based on color and shape information; (b) operating on the face candidates with a second algorithm, the second algorithm processing the face candidates using a posterior probability function classifier to determine the presence of the face; and wherein the first algorithm has a first rate of false positives, and the second algorithm has a second rate of false positives lower than the first rate of false positives."

Claim 5 is confirmed to be broader than the summary implied — it drops the color/shape and posterior-probability recitations entirely (quoted verbatim in Toshiba's amended complaint at the PTACTS record, and in Dr. Darrell's IPR declaration):

Claim 5. "(a) prescreening the image data with a first algorithm by using an algorithm interface adapter, the first algorithm determining a plurality of face candidates; and (b) operating on the face candidates with a second algorithm, the second algorithm processing the face candidates to determine the presence of the face; wherein the first algorithm has a first rate of false positives, and the second algorithm has a second rate of false positives lower than the first rate of false positives."

Claim 19 is the computer-program-product counterpart of claim 1 (storage medium storing a program performing steps (a)+(b), with the color/shape + false-positive limitations). Independents are 1, 5, 19; total 19 claims.

I also confirmed from the IPR record that the challenged claims in IPR2021-00330 were claims 1, 5, 10, 11 and 19 (Declaration of Dr. Trevor Darrell, https://ptacts.uspto.gov/ptacts/public-informations/petitions/[1544742](/patent/1544742)/). I was not able to verify the subject matter of claims 10–11 of the '294 patent, so I treat them generically below and flag them as unverified.

A. Framework and the effective-filing-date gate (do this first)

The '294 patent is a continuation of Ser. No. 09/514,436, filed 2000-02-28, now US 6,940,545. Its effective filing date is therefore 2000-02-28, and because that date precedes March 16, 2013, pre-AIA §§ 102/103 govern. This matters enormously, because a large fraction of the references printed on the face of the '294 patent are Kodak-family citations carried over from the parent, and several of them post-date the 2000-02-28 critical date.

Two drafting consequences of pre-AIA law that a challenger must respect:

1. § 103(c) / common ownership: Eastman Kodak was the original assignee of '294 and of US 5,432,863, US 4,503,508 and US 5,103,254. Because those three are § 102(b) art (issued long before 2/28/1999), § 103(c) offers the patentee no shelter. Only art qualifying solely under § 102(e) is disqualified by common ownership, and none of those three fall in that bucket.
2. § 102(e)-only art remains fair game against claim 1 here because none of Qian, Abdel-Mottaleb, Wang, Yamaguchi or Silverbrook was commonly owned with Kodak at the time of invention.

B. Claim-element map (what actually has to be taught or suggested)

Two observations that drive everything below:

• Claim 5 is the broadest and by far the weakest claim. It requires only (i) a two-stage arrangement, (ii) stage 2 operating on stage 1's candidates, and (iii) an FP-rate ordering. It does not require color/shape matching, does not require a posterior-probability classifier, and does not require a camera. Any two-stage face detector in which the first stage is tuned to over-detect reads on it. The FP-rate element is not a structural limitation at all — it is an inherent performance characteristic of any coarse-to-fine cascade, and inherency is not a source of patentable distinction (§ 211.05 MPEP; In re Best).
• Claim 1's "color and shape information" language is a verbatim characterization of Wu et al. Wu's abstract states the method builds "two fuzzy models to describe the skin color and hair color" and then compares extracted regions "with the prebuilt head-shape models by using a fuzzy theory based pattern-matching method to detect face candidates" (https://ieeexplore.ieee.org/abstract/document/[771326](/patent/771326); https://dl.acm.org/doi/abs/10.1109/34.771326). Color = skin/hair models; shape = 12×10 head-shape models; "face candidates" = the claim's exact noun.

C. Ground 1 — Wu (1999) + Schneiderman & Kanade (1998): the primary § 103 combination

This is the strongest and cleanest ground, because both references are already of record and both are marked "Examiner" on the face of the parent (see the citation table reproduced at https://patents.searchlight.law/doc/US6940545 and the PTO-892 listing on the '545 PDF).

What Wu teaches (claim 1 elements (a), and claim 5 element (a)):

• Fast, low-memory, color-based pattern recognizer scanning a window over a "complete range of scales and positions."
• The specification itself describes Wu as "a kind of ad-hoc pattern recognizer that searches for image windows that seem likely to contain faces based on color characteristics," and states that the algorithm "is very fast and requires very small amounts of both program memory and trained state" — both are advantages relevant to the camera-implementation claims.
• Critically, Wu's own output is denominated "face candidates," not face decisions. Wu does not purport to make a final determination.

What Schneiderman & Kanade teaches (claim 1 element (b), and claim 5 element (b)):

• "an algorithm for object recognition that explicitly models and estimates the posterior probability function, P(object|image)" (https://publications.ri.cmu.edu/probabilistic-modeling-of-local-appearance-and-spatial-relationships-for-object-recognition; http://www.cs.cmu.edu/~hws/publications.html).
• "We compute an estimate of P(object|image) in closed form… We have used this method for detecting human faces from frontal and profile views."
• Reported performance: 93.0% detection with 88 false alarms on a 125-image / 483-face test set. This is a disclosed, quantified low false-positive rate.
• The specification confirms the mapping: "the Component S implements a Bayesian classifier that performs maximum a posterior classification… the conditional probability distribution P(face|image)."

The motivation to combine — this is the crux, and it is unusually well supported:

1. Wu's genre label does the work. Wu explicitly calls its output candidates. A reference that characterizes its own output as provisional candidates inherently suggests a subsequent verification stage. This is not hindsight; it is on the face of the reference.
2. Perfectly reciprocal deficiencies. Wu is fast/low-memory but color-and-shape-based and therefore prone to false positives (its models are 12×10 low-resolution occupancy grids over a handful of poses). Schneiderman is accurate (93%, 88 false alarms) but its multi-resolution, subregion-anchored, 10⁶-bin evaluation over every window position and scale is computationally expensive — the spec concedes that component S "restrict[s] its more computationally intensive processing to the relatively few regions that have passed the first algorithm." Wu's paper and Schneiderman's paper therefore address the two halves of a single engineering problem, and each supplies what the other lacks.
3. The patent's own specification supplies the motivation in narrative form. The specification states the advantage verbatim: "the first component algorithm can be designed to operate quickly albeit with the potential for false positives and the second component algorithm can restrict its more computationally intensive processing to the relatively few regions that have passed the first algorithm." And, decisively, on the FP-rate element: the parameters of component W "are set quite low. The goal is for the component W to almost never miss a face, counting on further processing by algorithm S to eliminate the many false detections." That is an applicant-side statement of exactly why the first algorithm has the higher false-positive rate — i.e., a design rationale for the claimed FP-rate ordering. Admissions in the specification about the prior art and about the reason for a claimed parameter choice are binding on the patentee for obviousness purposes (Koninklijke Philips v. Google, 948 F.3d 1330, 1339 (Fed. Cir. 2020); PharmaStem v. ViaCell, 491 F.3d 1342, 1362 (Fed. Cir. 2007) — the latter formulation appears in a PTO Board decision retrieved at https://e-foia.uspto.gov/Foia/RetrievePdf?system=BPAI&flNm=fd2019005590-10-02-2020-0).
4. The stated problem is a design incentive, not a discovery. The specification's own background states: "The challenge is to implement these face detection methods reasonably in a camera with limited memory resources, and with low computational cost." Under KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007), a design incentive to solve a recognized problem with known elements, yielding predictable results, is sufficient motivation. A coarse-to-fine cascade is the textbook answer to "limited memory + low computational cost." Cf. In re Wilson (adapting a known technique to a new but analogous use is obvious).
5. The cascade was itself known in the face-detection literature of the period, which defeats any "unpredictable" or "new paradigm" argument. Schneiderman's own paper cites both Sung & Poggio, "Example-Based Learning for View-Based Human Face Detection" (whose distribution-based model relies on clustering non-face patterns to reject false positives) and Rowley, Baluja & Kanade, "Neural network-based face detection" (which arbitrates between multiple networks and uses a bootstrap procedure to harvest non-face examples). These citations are in the record via Schneiderman, and they establish that multi-stage/verification architectures were part of the ordinary artisan's toolkit. That converts the combination from "two references stapled together" into "known technique, known benefit."
6. Reasonable expectation of success. Both techniques were evaluated on overlapping benchmarks (Schneiderman reports results on combined MIT/CMU test sets), so the artisan had a concrete basis to expect that adding S behind W would cut false positives without destroying recall. S is a supervised classifier trained on face/non-face data; using it as a verifier for W's candidates changes only the input distribution, which is precisely the setting S was designed for.

Expected patentee response and rebuttal:

• "The references are non-analogous / from different sub-fields." Rebutted: both are face-detection references aimed at detecting faces in images. Same field of endeavor, same problem, same ordinary artisan.
• "The combination produces a new and unexpected result." The only "new" result is the FP-rate ordering, which is the predictable consequence of tuning W to high recall and putting a low-FP classifier behind it. KSR asks "whether the improvement is more than the predictable use of prior art elements according to their established functions" — here it is not.
• "Teaching away." I found no teaching away. There is no criticism of cascades or of color pre-screening in either reference; both are permissive. Note that the specification does not disparage Wu or Schneiderman — it endorses and incorporates both.
• "Non-obviousness of the specific (L,s,t) color metric and the low sigmoid parameters." These are unclaimed in claims 1/5/19. Unclaimed features cannot support non-obviousness of the claims. And the swap from Wu's Farnsworth space to (L,s,t) is an expressly acknowledged substitution ("Instead of using the Farnsworth perceptually uniform color space as suggested in the Wu et al. article, the present invention uses (L,s,t) color space") — a mere substitution of a known color metric, which is at most a § 103 design choice.

Conclusion on Ground 1: Claim 1, claim 5 and claim 19 are, in my assessment, rendered obvious by Wu in view of Schneiderman & Kanade. Claim 5 is the most vulnerable (it lacks color/shape and posterior-probability requirements entirely). Claim 1's added "algorithm interface adapter" limitation is a conventional bus/interface element the specification itself describes functionally ("for connecting various stored algorithms to the bus 40"); it adds essentially no patentable weight and, at most, is supplied by the standard equipment of any algorithm-loadable camera (e.g., a stored-algorithm digital camera per Lobo '616).

D. Ground 2 — Yow & Cipolla as newly-located primary art (not of record)

The IPR2021-00330 record shows Toshiba's expert, Dr. Trevor Darrell, applying Yow — i.e., Yow & Cipolla, Feature-based human face detection (Image and Vision Computing, 1997) — to the challenged claims, and characterizing it as follows (per the petition/declaration as retrieved at https://ptacts.uspto.gov/ptacts/public-informations/petitions/1544742/):

• "while the first stage of the method rejects grossly incorrect groupings of face candidates, it still produc[es]" errors — i.e., Yow's own paper discloses that stage 1 leaves residual false positives;
• stage 1 = "preattentive feature selection" — interest points selected by spatial filtering, edges linked by proximity/orientation/strength, a window box drawn around the edges;
• stage 2 = "attentive feature grouping" — features grouped into pairs, partial faces, then faces, with each grouping evaluated by Mahalanobis distance to a class mean against an admission threshold τ_j, using means and covariance matrices "define[d]" from training data.

Why this is dangerous for the patentee:

• Yow is not on the face of the '294 or '545 patent, so it was not before the examiner.
• Yow's two-stage structure with a first stage that over-includes and a second stage that validates is a direct factual predicate for the claim 5 FP-rate-ordered cascade — and arguably for anticipation of claim 5 under § 102(b) (publication 1997, more than one year pre-critical-date).
• For claim 1's "posterior probability function classifier," a Mahalanobis-distance class test is a Gaussian likelihood test; whether it reads on the claim's "posterior probability function classifier" is a claim-construction battleground. The patentee's best move is to argue for a narrow construction (a classifier outputting a posterior probability, or a MAP classifier). That defense fails against Ground 1, because Schneiderman is unambiguously a posterior-probability/MAP classifier and is already of record. So Yow is a supplementary ground, valuable mainly because it was not considered and because it independently corroborates the cascade.

Caveat: I retrieved only fragments of the Darrell declaration and petition; I could not confirm the full ground list, the secondary references, or the Board's institution outcome (the earlier sections flag the "Institution Denied" vs. "Pending – Instituted" conflict). Do not treat Ground 2 as verified beyond the quotations above.

E. Ground 3 — Alternative primary art for the two-stage + probabilistic-verification core

If a challenger needs to avoid Wu/Schneiderman (e.g., to blunt a "the examiner already considered these" narrative), the following of-record references supply the same elements:

The Moghaddam '833 reference is notable: it is a probabilistic (Bayesian eigenspace) object detector issued January 1998, i.e., § 102(b) art, and it squarely discloses computing a class-conditional probability for a face/non-face decision. Combined with Bang '325 or Kung '470 as the candidate-generation stage, this yields a Wu/Schneiderman-free § 103 ground reaching claim 1.

F. Ground 4 — the camera-function dependent claims (near-certain obviousness)

The '294's dependent claims largely recite applications of a detected face to ordinary camera functions, each of which has a dedicated reference of record. These are the easiest § 103 holdings in the patent:

Composition claims caveat worth stating fairly. A patentee would argue that keying composition advice to a detected human face's centroid (rather than to a generic subject region or user-selected point) is a non-obvious application. That is the only genuinely arguable point in the dependent set — but it is thin: US 5,873,007 already automates composition guidance, and the specification itself frames the face only as evidence of where "the main subject" is, which is exactly what a composition-guidance system needs. The combination is a predictable application of two known elements.

G. Claim-by-claim vulnerability assessment

H. Secondary considerations and defenses (a candid assessment)

• No objective evidence of non-obviousness appears in the record I could access. The prior sections document an extensive Monument Peak Ventures assertion program (eight-plus defendants; the 2019 Toshiba action; IPR2021-00330; CAFC 21-1052). Licensing/assertion activity is weak secondary-consideration evidence absent a demonstrated nexus between the claimed invention and the licensing demand — and here the district court held the asserted claims patent-ineligible under § 101 at the pleading stage (Aug. 12, 2020; final judgment Sept. 11, 2020), which undercuts any narrative that the market rewarded the claimed advance. I found no evidence of commercial success, copying, industry praise, or a long-felt unmet need attributable to the claims.
• § 101 may moot § 103 in the Toshiba dispute. If the Federal Circuit affirmed the § 101 holding in 21-1052 (an outcome the prior sections could not verify), the claims fell on eligibility grounds and the obviousness question was never authoritatively decided. A § 103 analysis remains independently relevant only for other defendants and for the IPR.
• Claim-differentiation argument the patentee will press. The patentee's best § 103 defense is that Wu's fuzzy pattern-matching detector and Schneiderman's Bayesian detector are complete, standalone detectors, so combining them is "hindsight" — a facial "why would you need two?" argument. This fails because Wu itself denominates its output as "candidates," because the specification concedes the speed/cost-versus-accuracy tradeoff as the invention's entire advantage, and because the face-detection literature of the period (Sung & Poggio; Rowley/Baluja/Kanade, both cited within Schneiderman) establishes multi-stage verification as known and beneficial.

I. Limits of this analysis — what a full § 103 opinion still needs

1. Claims 10 and 11 of the '294 patent are unverified. The earlier summary's description of the claim set (claims 10–18 depending from claim 5, with camera-implemented and red-eye variants) is an inference from the parent '545 claim set and does not match the verified fact that '294 claim 1 itself recites the FP-rate limitation. Confirm claims 6–19 verbatim from PatentCenter before relying on any per-claim conclusion above.
2. I could not verify the ground(s) actually presented in IPR2021-00330 beyond the Darrell declaration's identification of the challenged claims and its treatment of Yow; nor could I resolve the institution-outcome conflict ("Institution Denied" per GreyB/IPVerse vs. "Pending – Instituted" per Google Patents), nor the disposition of CAFC 21-1052.
3. I did not run a full, independent prior-art search. The references above are (a) the references of record on the '294/'545 family and (b) the Yow reference surfaced through the IPR record. A complete § 103 opinion would run a classification search across G06K9/00 (the '294's original class), G06V40/16x, and H04N23/61x, and would specifically verify the two-stage character of US 5,715,325 (Bang), US 5,642,431 (Poggio) and US 5,850,470 (Kung) against their specifications, and the bibliographic identity of US 5,873,007 ("Picture Composition Guidance System").
4. Do not confuse JP 7583294 B2 (Shinagawa Refractories, refractory-brick composition) with US 7,583,294.

Bottom line. On the record of the '294 patent itself, the claims are highly vulnerable under § 103. The core independent claims 1, 5 and 19 are rendered obvious by the combination of Wu, Chen & Yachida (1999) — which supplies the fast, color-and-shape, window-based face-candidate prescreener — with Schneiderman & Kanade (CVPR 1998) — which supplies the posterior probability function classifier with a disclosed low false-alarm rate — the motivation being the expressed in-camera memory/throughput constraint, the references' mutually complementary error characteristics, Wu's own designation of its output as "candidates," and the applicant's binding specification statements that component W is tuned to "almost never miss a face, counting on further processing by algorithm S to eliminate the many false detections." Claim 5, which omits color/shape and posterior-probability requirements entirely and reduces the invention to an FP-rate-ordered cascade, is additionally at anticipation risk from Yow & Cipolla (1997), art that was not before the examiner. The dependent claims fall to the ordinary combination of that core with the camera-function references the applicant itself cites — Bell '254, Tomitaka '809, Kinjo '752, Narisawa '000, Benoni/Brooks '508, Benati '863, Silverbrook '817, Yamaguchi '914 — and, for the composition claims, with Grill & Scanlon (1990) and US 5,873,007.