US 10904539

I'll search for litigation and docket information on this specific patent, then summarize.

Let me search for the reexamination proceeding and any 2026 litigation involving this patent.

Let me check for any 2026 Federal Circuit appeal involving this patent and the reexamination number.

US Patent 10,904,539 B2 — Summary

Bottom line: I found no Federal Circuit docket in 2026 (or any year) specifically involving US 10,904,539. The 2026 CAFC activity I located involving Ideahub concerns different patents (see "Litigation" below). The most significant 2025–2026 event for the ’539 patent is an ex parte reexamination filed by Unified Patents, which ended with the original claims confirmed and new claims added.

Bibliographic data (per the authoritative patent record)

Source: https://patents.google.com/patent/US10904539/en

Abstract (verbatim)

"Disclosed are methods, apparatuses, and systems for encoding and decoding an image. The present invention provides an intra prediction unit receives an input image, removes high frequency ingredients by low pass filtering an encoded luma pixel value in the input image during intra prediction, and generates a prediction block by predicting a chroma pixel value by using a low pass filter (LPF) LM chroma mode for applying an LM chroma mode, which is an extended chroma mode technique for generating a prediction block by predicting the chroma pixel value by applying a correlation between color planes to the luma pixel value having removed therefrom the high frequency ingredients."

Plain-language overview of the independent claims

The five issued claims track the specification's two-mode scheme: a conventional "LM chroma mode" (predict chroma from reconstructed luma via α·G + β) and a "low pass filter (LPF) LM chroma mode" (same, but with the luma plane low-pass filtered first to strip high-frequency content where RGB color-plane correlations are weak).

Claim 1 — Image encoding apparatus (processor functioning as an intra predictor):

• Determines whether an "LMmode" (chroma predicted from luma) applies to the current block.
• If applied, selects one of several LMmode candidates and predicts the chroma pixel with the selected candidate.
• A flag indicating LMmode application is encoded into the bitstream; when LMmode applies, index information identifying the selected candidate is additionally encoded.
• Candidates include a first LMmode and a second LMmode.
• Chroma prediction uses a first filtered luma value, a weight (α), and an offset (β). Under the first LMmode, the weight and offset are obtained based on a second filtered luma value; under the second LMmode, they are obtained without the second filtered luma value.
• Both a horizontal neighboring luma pixel and a vertical neighboring luma pixel adjacent to the co-located luma pixel are relied on to derive the first filtered luma value, which is a weighted sum of the co-located luma pixel plus those horizontal and vertical neighbors.
• Prediction equation recited: C_pred[x,y] = α·G_recon_lpf[x,y] + β, with the second filtered luma value obtained from a neighboring luma pixel.

Claim 3 — Image decoding apparatus — Mirror image of claim 1: the LMmode flag is parsed from the bitstream, index information for the selected LMmode candidate is parsed when LMmode applies, and the weight/offset are obtained based on a linear relationship between luma and chroma pixels. Otherwise the same first/second-filtered-luma-value and weighted-sum framework applies. (Note: the record's claim text contains the literal typo "neighboring lima pixel.")

Claim 4 — Image encoding method — Method-step counterpart to claim 1 (determine → select → predict), with explicit sub-steps of obtaining the first filtered luma value, obtaining the weight and offset from the luma/chroma linear relationship, and predicting the chroma pixel with α·G_recon_lpf[x,y] + β.

Claim 5 — Image decoding method — Method-step counterpart to claim 3, with the flag and index parsed from the bitstream.

Claim 2 (dependent) — Specifies that the horizontal neighboring luma pixel is at least one of a left or right neighbor, and the vertical neighboring luma pixel is at least one of a top or bottom neighbor.

Practical takeaway: Claims 1/3 (apparatus) and 4/5 (method) protect an encoder/decoder that signals, and then selectively applies, one of multiple LM (luma-based chroma prediction) modes — including a filtered variant — using a flag plus index, with the chroma prediction computed as a weighted, offset linear function of a spatially-filtered luma value derived from a co-located pixel and its immediate vertical/horizontal neighbors.

Prosecution, reexamination, and standards status

• Spec-level filters: The specification's Equation 3 gives an example low-pass kernel as G_lpf[x,y] = 0.5·G[x,y] + 0.125·G[x+1,y] + 0.125·G[x−1,y] + 0.125·G[x,y+1] + 0.125·G[x,y−1]. ⚠️ Literal-text caution: the "Technical Solution" section of the record separately renders this as "...+0.125 G_recon[x,y−1]+0.125 G_recon[x,y−1]" (the y−1 term appears duplicated rather than showing the y+1 term). I am reporting the text literally rather than correcting it.
• Ex parte reexamination: On 2025-01-31, Unified Patents filed an ex parte reexamination against US 10,904,539 (control number shown by Unified as 90/019,837). The Central Reexamination Unit granted the request on 2025-03-07, finding substantial new questions of patentability on all challenged claims. Unified was represented by Dagim Tilahun and Jon Bowser (Haynes and Boone) and in-house counsel Jessica L.A. Marks and Michelle Aspen. Sources: https://www.unifiedpatents.com/insights/2025/3/7/ideahub-video-codec-patent-challenge-instituted and https://www.unifiedpatents.com/insights/tag/Ideahub
• Reexamination outcome (2026): The patent record's legal events show a reexamination decision dated 2026-02-03 (kind code C1): "THE PATENTABILITY OF CLAIMS 1-5 IS CONFIRMED. NEW CLAIMS 6-20 ARE ADDED AND DETERMINED TO BE PATENTABLE." (Effective date listed as 2026-01-26.) I do not have the text of new claims 6–20, so I cannot characterize their scope.
• Standards: The ’539 patent has been designated as essential to the VVC standard in the Access Advance VVC patent pool.

Litigation / CAFC check (with explicit uncertainty)

• No CAFC appeal of this patent found. My searches did not surface any Federal Circuit 2026 docket for US 10,904,539.
• The 2026 CAFC matters involving Ideahub that I did find concern other patents and should not be attributed to the ’539 patent:
  • IdeaHub Inc. v. Unified Patents, LLC, No. 2024-1684 (Fed. Cir. Apr. 10, 2026) — nonprecedential, affirmed the PTAB's PGR2022-00044 decision on US 11,122,274. https://law.justia.com/cases/federal/appellate-courts/cafc/24-1684/24-1684-2026-04-10.html
  • IdeaHub Inc. v. Unified Patents, LLC, No. 2022-1160 (Fed. Cir. Feb. 10, 2023) — earlier affirmance concerning US 9,641,849.
• Ideahub's 2025–2026 district court/ITC campaign (via subsidiary Pantech Corp against Lenovo, TCL, HMD, OnePlus, Tinno) asserted LTE/5G patents (US 9,548,839; 11,051,344; 11,659,503; 12,267,876) — not the ’539 patent. Those cases settled in mid-2026.

Uncertainty note: I could not independently verify a district court or ITC complaint asserting the ’539 patent, and the reexamination certificate number/date details above come from the Google Patents legal-events record rather than a directly retrieved USPTO C1 certificate document. If you need the certificate PDF or the text of new claims 6–20 for claim-charting purposes, that should be pulled directly from USPTO PatentCenter for application 15/026,154 and reexamination control 90/019,837.

I'll research litigation involving US patent 10904539 across patent litigation databases.

The first results referenced U.S. Patent 10,949,339 (a Netlist memory patent) — that is a different patent number from 10,904,539, so I must not conflate them. Let me search specifically for the correct number and its owner.

Found key facts. Let me dig deeper into district court litigation and the reexamination details.

Litigation Involving U.S. Patent No. 10,904,539

Important identification note first: In searching for this patent I encountered extensive results for U.S. Patent No. 10,949,339 (a Netlist, Inc. memory-module patent asserted against Samsung, e.g., Netlist v. Samsung, No. 2:21-cv-00463 (E.D. Tex.), and IPR2022-00639). That is a different patent number (10,949,339 ≠ 10,904,539) with a different owner and technology. I have excluded those results per your instruction not to return similar numbers. Everything below concerns 10,904,539 only.

The patent at issue

• Patent: U.S. 10,904,539 B2 — "Method, device and system for encoding and decoding image"
• Application: US 15/026,154; PCT/KR2014/009221; priority to KR 10-2013-0116220 (2013-09-30); granted 2021-01-26
• Inventors: Yoon Sik Choe; Jang Won Choi
• Original assignee: University-Industry Foundation (UIF), Yonsei University → current assignee: Ideahub Inc. (assignment recorded 2021-05-26)
• Subject matter: Intra-prediction using LM (linear model) chroma modes, including a low-pass-filtered LM chroma mode ("LPF LM chroma mode") that predicts a chroma pixel from a filtered luma value. Designated as essential to the VVC (H.266) standard in Access Advance's VVC patent pool.
• Source: https://patents.google.com/patent/[US10904539](/patent/US10904539)/en ; https://www.unifiedpatents.com/insights/2025/2/3/ideahub-video-codec-patent-challenged

Known proceedings specifically involving the '539 patent

1. Ex parte reexamination — Unified Patents, LLC v. Ideahub Inc. (USPTO, not district-court litigation)

• Type: Ex parte reexamination (USPTO administrative proceeding — not litigation)
• Plaintiff/Petitioner (requester): Unified Patents, LLC
• Patent owner: Ideahub Inc.
• Jurisdiction: U.S. Patent and Trademark Office
• Filed: January 31, 2025
• Status/outcome: Reexamination Certificate C1 issued (decision dated Feb. 3, 2026, effective Jan. 26, 2026) — the original claims 1–5 were confirmed patentable, and new claims 6–20 were added and held patentable.
• Sources:
  • Unified Patents news post: https://www.unifiedpatents.com/insights/2025/2/3/ideahub-video-codec-patent-challenged (Unified represented by Dagim Tilahun and Jon Bowser of Haynes and Boone, plus in-house counsel Jessica L.A. Marks and Michelle Aspen)
  • Google Patents legal events for US10904539B2 (RR "Request for reexamination filed," effective 2025-01-31; LIMR "Reexamination decision… Kind code C1… CLAIMS 1-5 CONFIRMED. NEW CLAIMS 6-20 ADDED AND DETERMINED TO BE PATENTABLE"), https://patents.google.com/patent/US10904539/en

2. District-court litigation asserting the '539 patent

• None found. I could not identify any federal district court (or ITC, or foreign court) case in which U.S. 10,904,539 is an asserted patent. Based on the available search coverage, the only known adversarial proceeding against the '539 patent is the Unified Patents ex parte reexamination described above.

Related Ideahub activity — NOT involving the '539 patent (flagged to prevent confusion)

Ideahub is an active NPE, but its known litigation asserts other patents (DASH/MPEG-DASH streaming patents and cellular standard patents), not the '539 patent:

• Helios Streaming, LLC & Ideahub, Inc. v. Vudu, Inc., No. 1:19-cv-01792 (D. Del., filed 2019-09-24) — asserted eleven MPEG-DASH patents (e.g., '736, '830, '660, '414, '145, '130, '373, '562, '805, '558, '493). Not the '539 patent.
• Helios Streaming v. Showtime Digital/Showtime Networks (D. Del.) and Helios Streaming v. Starz Entertainment (C.D. Cal., No. 19-2140) — same DASH portfolio. Not the '539 patent.
• Helios Streaming (an Ideahub company) v. Rakuten and DAZN — UPC and Munich I Regional Court (DASH patents). Not the '539 patent.
• Pantech Corporation (Ideahub affiliate) v. HMD Global OY, No. 5:25-cv-00092 (E.D. Tex., filed 2025-07-03), and related suits against Lenovo, TCL, Tinno Mobile, and OnePlus — asserted patents are described as 5G/LTE standard-essential cellular patents. I could not confirm that the '539 patent is among them.
• IdeaHub Inc. v. Unified Patents, LLC, No. 24-1684 (Fed. Cir.) — an appeal concerning a different patent (intra-mode/written-description issues). Not the '539 patent.

Bottom line / confidence

• Confirmed: U.S. 10,904,539 was the target of a Unified Patents ex parte reexamination filed Jan. 31, 2025, resulting in Reexamination Certificate C1 (claims 1–5 confirmed; new claims 6–20 added, Feb. 2026).
• No district-court or ITC litigation asserting U.S. 10,904,539 was found. I state this with moderate confidence given search coverage limits.
• Caveat: My searches were constrained, and PACER / the Unified Patents litigation portal (https://portal.unifiedpatents.com/litigation/caselist) may reveal newly filed complaints asserting the '539 patent — particularly given it is a VVC standard-essential patent in a pool that is actively being enforced. A direct PACER docket-name search for "Ideahub" and a query of the Unified Patents litigation database are recommended to confirm the negative on district-court cases. I do not have high confidence that no such case exists; I can only report that none surfaced.

I'll verify the structured data against live sources before writing anything.

The first result is a trap: 10,949,339 (Netlist memory patent) is a different patent from 10,904,539. Let me search precisely for the '539 patent and its owner.

Key finding emerging: no AIA trial on the '539 — but there's a Unified Patents ex parte reexam. Let me verify family members and any litigation/PTAB activity.

Proceedings overview

Zero AIA trial proceedings (IPR / PGR / CBM) on file for U.S. 10,904,539. Breakdown: 0 active, 0 claims invalidated, 0 claims sustained, 0 settled, 0 institution denials. The USPTO ODP structured list is empty and independent web searching surfaced no IPR, PGR, or CBM petition naming the '539 patent as of 2026-09-29. The only post-issuance adversarial event is an ex parte reexamination — which is a prosecution proceeding, not an AIA trial — and it confirmed claims 1-5 and added new claims 6-20. Bottom line for a defendant: there is no PTAB knock-out opinion to lean on, and unlike the framing in the prompt ("survived two IPRs"), the accurate posture is "never IPR'd, validity-scrubbed in reexam, and now carrying more claims than it issued with." That is a hardening event, not a weakening one, and it means an invalidity case must be built from scratch.

⚠️ Identifier trap — do not confuse these. A large volume of PTAB material references U.S. Patent 10,949,339 (Netlist, memory buffers): IPR2022-00639, IPR2023-00204, PGR2025-00071, PGR2026-00001, and Netlist v. Micron, No. 2024-1707 (Fed. Cir. 2026-09-02). That is a different patent, different owner, different technology. None of it touches 10,904,539. Likewise, the extensive Ideahub PTAB docket below is on other Ideahub patents, not this one.

No AIA trial proceedings — patent is PTAB-virgin

• Type: n/a (no petition ever filed)
• Filed: n/a
• Status: No proceedings docketed; ODP returns none, and no petition was surfaced by web search
• Judge panel: n/a
• Petition grounds: n/a
• Institution decision: n/a
• Final Written Decision: none issued
• Settlement / termination: n/a
• Appeal: none from this patent
• Defensive value: Because no IPR/PGR was filed, no § 315(e)(2) estoppel attaches to any party — a defendant today has a completely unencumbered prior-art menu and can raise any § 102/§ 103/§ 112 ground it can find. The trade-off: no FWD, no institution decision, and no claim-construction ruling exist to borrow. Any invalidity theory is greenfield work. Note also that claims 1-5 have already been examined again by an examiner and confirmed, so a petitioner must beat a record that has now survived two Office reviews.

Adjacent activity — not an AIA trial (included because it is the only real post-issuance proceeding)

Ex parte reexamination — Unified Patents, LLC (third-party requester) v. Ideahub Inc.

• Type: Ex parte reexamination (35 U.S.C. §§ 301–307) — expressly not an IPR, PGR, or CBM
• Filed: 2025-01-31 (per Unified Patents; the patent's legal events show the reexam request docketed 2025-04-01 with an effective date of 2025-01-31)
• Status (verbatim from Google Patents legal events, 2026-02-03): "REEXAMINATION CERTIFICATE; THE PATENTABILITY OF CLAIMS 1-5 IS CONFIRMED. NEW CLAIMS 6-20 ARE ADDED AND DETERMINED TO BE PATENTABLE." Plain English: IdeaHub won. The existing claims came through intact and the owner walked away with fifteen additional claims.
• Judge panel: n/a (examiner corps, not APJs)
• Grounds: I could not verify the specific references or statutory basis — ex parte reexam files are not public in the structured ODP feed I was given, and I will not invent them.
• Control number: not verified. I could not confirm the 90/xxx,xxx control number from the sources available; treat it as unconfirmed rather than assume one.
• Settlement / termination: n/a — reexam is ex parte; requester identity may be shielded under § 302, and Unified publicly self-identified anyway.
• Appeal: none identified.
• Defensive value: This is the worst possible outcome for a defendant hoping to free-ride: it re-validated claims 1-5 against a professional NPE-defense aggregator and enlarged the claim set. Note the new claims 6-20 are untested claims with no prosecution history a defendant has seen — I could not verify their text, so the very first thing to do is pull the reexam certificate and the added claims. Also relevant context: Unified describes the '539 as declared essential to the VVC standard and a member of the Access Advance VVC patent pool, while Unified's product is Free Ad-Supported Streaming TV (FAST) / the "SEP Video Codec Zone." That is a SEP-licensing dispute dressed as a validity fight.

Strategic summary

Claim status: claims 1-5 SUSTAINED (reexam-confirmed, never IPR-tested); claims 6-20 UNTESTED (newly added). No claim of the '539 has ever been canceled by the PTAB or by the examiner. Claim 1 as issued recites the LMmode/plural-candidates architecture — a first LMmode and a second LMmode, a flag for whether the LMmode applies, index information selecting among LMmode candidates, and prediction by Cpred[x,y]=αGrecon_lpf[x,y]+β using a filtered luma value derived from the co-located pixel plus horizontal and vertical neighbors (see claim 1, which recites a weighted-sum operation over the co-located, horizontal-neighboring and vertical-neighboring luma pixels). Any infringement theory, whether it points at LM chroma or LPF-LM chroma, sits on those claims — and they are alive.

Estoppel landscape — a clean slate, with one § 325(d) caveat. Because no IPR/PGR was ever filed against the '539, no one is barred by § 315(e)(2). A defendant can raise every ground it can find, including art that was before the examiner in the reexam. The real friction is § 325(d) discretion: if a defendant's best art is the same art the examiner already considered and rejected during reexamination, the Board may deny institution as "previously presented to the Office." The practical move is to find different art and a different framing. Compare the sibling-patent history: in IPR2020-00702 ('849) and PGR2022-00044 ('274), Unified won on § 112 written description rather than § 103 — an instructive lesson that the productive attack surface for this family has been the specification, not the references.

Pattern signals. The patent owner, Ideahub Inc. (Korean NPE; moved from Yonsei University's University Industry Foundation, assigned to Ideahub 2021-05-14, reels 038879/0223 and 056358/0816), is a serial and aggressive PTAB appellant: IPR2020-00702 ('849) → appealed as Ideahub v. Unified Patents, No. 2022-1160, affirmed 2023-02-10; PGR2022-00044 ('274) → appealed as No. 2024-1684, argued 2026, on whether collateral estoppel can rest on an unconstrued claim term. That appeal lost — all challenged claims were held unpatentable for lack of written description. Ideahub has also litigated heavily on its ETRI-sourced streaming patents (Helios Streaming v. Starz, Vudu, Crackle, Showtime), where Vudu's IPRs (IPR2020-01688 on the '830, IPR2020-01689 on the '414) produced a complete victory — all challenged claims unpatentable and the motion to amend denied. Defensive aggregator: yes — Unified Patents is the recurring petitioner across this owner's portfolio and, notably, is the reexam requester here. Unified's counsel of record in the '539 reexam: Dagim Tilahun and Jon Bowser of Haynes and Boone, with in-house counsel Jessica L.A. Marks and Michelle Aspen.

Family depth is the real problem. The '539 is one node in a five-member U.S. continuation family all claiming the 2013-09-30 priority date: US 10,904,539 (this patent), US 11,254,221 (filed 2020-12-30), US 11,683,506 (filed 2022-01-14), US 12,301,842 (filed 2023-06-19), and US 2025/0267293 A1 (filed 2025-04-11, still pending). A sixth-related sibling, US 11,425,419 ("…using LM chroma prediction") covers the correlation-threshold variant of the same idea. A well-aimed hit on the '539 does not retire the family — the owner will simply assert the next continuation. Adjusted expiration on this patent runs to 2035-06-10. Any defense strategy must be family-wide, not patent-by-patent.

Recommended next steps

1. Correct the record internally before anything else. Build the docket from the reexam, not from an IPR. Pull the reexamination certificate and the text of new claims 6-20 at USPTO PatentCenter for U.S. 10,904,539. The new claims are the widest, freshest, and least-defended part of the patent, and I could not verify their scope from the available sources. If a demand letter cites "claims 1-5," those claims were expressly confirmed patentable on 2026-02-03 — there is no cancellation to point to, and arguing otherwise is a misstatement of the record.
2. Don't confuse this with the '339. Stripe out any internal reference to IPR2022-00639, IPR2023-00204, PGR2025-00071, or PGR2026-00001; all four concern U.S. 10,949,339 (Netlist), not 10,904,539. Mis-citation here would be embarrassing at the Board.
3. No AIA clocks are running — there is no 1-year § 316(a)(11) trial deadline, no institution-decision deadline, and no briefing schedule to track. The only live clock is district-court/ITC § 315(b) one-year bars if and when you are served.
4. If you file, file on § 112 first. The sister-patent record is unambiguous: Unified's wins on the '849 and '274 were written-description holdings under § 112, and the Federal Circuit affirmed the '849 finding. A § 103 attack on the '539 must get past a reexam record, and the family's shared specification is where the demonstrated vulnerability lies.
5. Budget for a family-wide fight and a SEP overlay. Five U.S. family members, one still pending as of 2025-04-11, plus an Access Advance VVC-pool essentiality declaration, means a validity win on the '539 alone is not a resolution. Explore whether the dispute is really a pool FRAND/royalty-rate question, and check § 325(d) exposure before selecting art — the reexam examiner already rejected something, and reusing that art invites discretionary denial.
6. Monitoring. Set an alert for new PTAB filings naming 10,904,539 and its four continuations. Given Unified's portfolio-wide campaign and the VVC pool's assertion posture, a petition against this patent (or a sibling) would not be surprising — but as of today, none exists, and I found none in web searches either.

Uncertainty disclosure: I could not verify (a) the reexamination control number, (b) the specific prior art or statutory grounds Unified relied on in the reexam, (c) the text of new claims 6-20, or (d) any district-court or ITC action asserting the '539 itself. I searched for all four and found nothing definitive; I have not filled those gaps by inference.

I'll research the assignment chain, IdeaHub's litigation history, and the reexamination record before building the report.

Key new leads: an Ariscale→IdeaHub assignment, Helios Streaming as exclusive licensee, and the reexam outcome. Let me dig further.

Ownership Analysis — US 10,904,539 B2

"Method, device and system for encoding and decoding image" · Appl. 15/026,154 · PCT/KR2014/009221 · Priority 2013‑09‑30 (KR 10‑2013‑0116220) · Issued 2021‑01‑26 · Current assignee of record: Ideahub Inc. (Seoul, KR)

Inventors

Pattern notes

• Both inventors are on a single assignment instrument to Yonsei's tech‑transfer foundation, executed ~30 months after the KR priority filing. That lag coincides with US national‑stage entry of PCT/KR2014/009221 (app. 15/026,154) — i.e., it is a procedural formality of the 371 filing, not evidence that the inventors disengaged or that the rights were being warehoused.
• No evidence of inventors transferring to another entity, departing, or of a fire‑sale triggered by inventor exit. There is also a sibling family member with the same Yonsei assignee (KR 10‑1530782, priority 2013‑12‑03), which is consistent with a normal university research group, not a discarded portfolio.
• I could not verify a corporate R&D employer for either inventor from the sources available; treat the "Yonsei‑affiliated" conclusion as inference from the assignment, not a documented employment record.

Original assignee

University‑Industry Foundation (UIF), Yonsei University (a/k/a "University Industry Foundation of Yonsei University"), Seoul, KR.

• Business: university technology‑transfer / IP‑management foundation for Yonsei University. It is not a product company.
• Shipped a product embodying the claims? No. There is no evidence of any encoder/decoder product or service. The claims cover an intra‑prediction chroma/ECM encoding scheme; the commercial relevance is standardization (HEVC Range Extensions → VVC), which is a licensing, not a product, channel.
• Current status: operating as an entity (it acted as assignor of record in 2021 and as predecessor‑in‑interest in the 2021 recording), but it no longer owns this patent.

Current owner of record: Ideahub Inc., 7 Heolleungro, Seocho‑gu, Seoul 06792, Republic of Korea. Ideahub is a Korean patent‑monetization company with no evident product line; Unified Patents describes it in its own filings/publications as "Ideahub Inc., an NPE" (Unified Patents — IdeaHub). It acquires portfolios (e.g., the Ariscale Inc. → Ideahub Inc. patent purchase, assignment document circulated/recorded 2018, covering Ariscale's intra‑prediction and streaming families) and monetizes through (a) an exclusive‑licensee vehicle for litigation — Helios Streaming, LLC, a Delaware LLC at 9880 Irvine Center Drive, Suite 200, Irvine, CA, which "holds the exclusive right to assert all causes of action" — and (b) the Access Advance VVC patent pool, in which the '539 patent has been declared essential.

Assignment timeline

Provenance note: In this session I could not load the Assignment Center record pages directly (they are JavaScript‑rendered); the reel/frame data below is transcribed from the USPTO assignment abstracts surfaced in the patent's legal‑events record (Google Patents, which mirrors Assignment Center). Two recorded assignments exist for this patent. Correspondent fields could not be retrieved and are the main verification gap — see Signal 3.

• 2016‑03‑29 / 2016‑04‑20 (executed) / recorded 2016‑06‑10 — Reel 038879/0223

  • Conveyance: Assignment (inventor → university foundation)
  • Assignor: Choi, Jang Won and Choe, Yoon Sik
  • Assignee: University‑Industry Foundation (UIF), Yonsei University
  • Correspondent: Not retrievable in this session.
  • Context: Standard prerequisite perfecting of title at PCT national‑stage entry; not a transfer of control.

• 2021‑05‑14 (effective/executed) / recorded 2021‑05‑26 — Reel 056358/0816

  • Conveyance: Assignment (sale/transfer of entire interest)
  • Assignor: University Industry Foundation of Yonsei University
  • Assignee: IDEAHUB INC.
  • Correspondent: Not retrievable in this session. Lead to check: for the same era of IdeaHub filings the recorded prosecution correspondent is Kyeongsu Im, in‑house, ks.im@ideahub.co.kr, 9 Heolleungro Ste 909, Seocho‑gu, Seoul 06792 (from the IdeaHub/Ariscale filing papers). Whether Im also signed the 056358/0816 cover sheet must be confirmed on the Assignment Center record — do not treat it as established.
  • Context: Transfer to a monetization entity. Issued 2021‑01‑26; sold ~3.5 months post‑issuance to a company with no products, immediately before IdeaHub's SEP‑licensing campaign (VVC pool) and its continuation‑filing program on this family.

• Post‑2021 (no recorded assignment to a third party): ownership remained with Ideahub Inc. Ideahub prosecuted four continuation applications from this family, all still Ideahub‑owned: US 17/138,076 → US 11,252,421; US 17/576,610 → US 11,683,506; US 18/337,356 → US 12,301,842; US 19/177,220 → US 2025/0267293 A1 (pending). This is evidence of sustained investment in the family, not abandonment.

• 2025‑01‑31: Unified Patents filed an ex parte reexamination of the '539 patent (control no. 90/019,837); instituted 2025‑03‑07. 2026‑02‑03: Reexamination Certificate C1 issued — claims 1‑5 confirmed patentable, new claims 6‑20 added. Ideahub remains owner of record and participated as patent owner, which independently corroborates continuous ownership through 2026‑01‑26.

There are Assignment Center records for this patent (two). Only the second is a change of beneficial owner.

Timeline diagram

timeline
    title Ownership of US 10904539
    2013 : KR priority application filed
    2014 : PCT filed by Yonsei UIF
    2016 : Inventors assign to Yonsei UIF
    2021 : Patent issued 26 January
         : Yonsei UIF sells to Ideahub Inc
    2025 : Unified files ex parte reexam
         : Reexam instituted by CRU
    2026 : Certificate C1 confirms claims 1 to 5

NPE / troll-pattern signals

1. Shell‑entity transfer — PRESENT (substance), naming tells absent.
   The patent moved from a non‑practicing university foundation to IDEAHUB INC. via reel 056358/0816, effective 2021‑05‑14, ~15 weeks after issuance. Ideahub holds no product line practicing the claims; it is a portfolio acquirer (Ariscale purchase, 2018) and licensor. However, the classic naming/address tells are absent: no "IP/Holdings/Ventures" suffix, no Delaware or Texas registered‑agent address, no anonymous single‑member LLC — Ideahub is a real Seoul corporation with in‑house IP staff (correspondent Kyeongsu Im, ks.im@ideahub.co.kr). The "shell" characterization therefore rests on conduct (buy‑and‑license), not the name.

2. Known asserter in the chain — PRESENT (not on the legacy lists).
   Current assignee Ideahub Inc. is expressly identified as an NPE by Unified Patents, which has challenged multiple Ideahub patents (Unified Patents — IdeaHub). Ideahub is a repeat plaintiff through its exclusive licensee: Helios Streaming, LLC + Ideahub, Inc. sued Vudu (D. Del. 1‑19‑cv‑01792), Crackle (1‑19‑cv‑01818), Showtime Digital (1‑19‑cv‑01978) and Starz (C.D. Cal. 8‑19‑cv‑02140). It does not match Acacia, Marathon, IV, Wi‑LAN/Conversant, Vringo, Pendrell, Round Rock, etc. The '539 itself is declared VVC‑essential in Access Advance's pool and was targeted by Unified's ex parte reexam — the signature of an asserted SEP. Caveat: I found no district‑court complaint that names the '539 patent specifically.

3. Repeat correspondent across the chain — UNCLEAR (verification gap).
   I could not retrieve the recorded correspondent for either reel (038879/0223 or 056358/0816) in this session, so I cannot confirm or deny a repeat attorney. This is the single highest‑value item to pull from Assignment Center. Concrete leads to test: (i) does Kyeongsu Im (IdeaHub in‑house, Seoul) appear as correspondent on 056358/0816? (ii) does the same correspondent/firm appear on both the Yonsei→Ideahub recording and the Ariscale→Ideahub recordings? If one name recurs across those, that upgrades Signal 3 to present and is the classic "the LLCs change, the lawyer doesn't" tell.

4. Cascading transfers — NOT PRESENT.
   Only two recorded links in 13 years, ~5 years apart, ending in a single owner that has held the patent for 5+ years and is still prosecuting continuations from it (US 11,252,421; US 11,683,506; US 12,301,842; US 2025/0267293 A1). There is no chain of chained LLCs, no common address across assignees, and no rapid flipping.

5. Pre‑litigation transfer — NOT PRESENT / UNCLEAR.
   The transfer to Ideahub (2021‑05‑14) is not within 6 months of any suit I can find naming the '539. The Helios‑backed suits (Sept 2019) predate the acquisition and assert entirely different patents (ETRI/MPEG‑DASH streaming family). The only '539‑specific adverse action found is a 2025 ex parte reexamination, which is triggered by a challenger, not by the patent owner's filing of suit. Marked not established because assertion of this specific patent may have occurred outside public court filings (e.g., pool licensing demands).

6. Bankruptcy fire‑sale — NOT PRESENT.
   Transferor is a university foundation; no Chapter 7/11, no trustee sale, no court‑approved asset sale is implicated in either recording.

7. Privateering — NOT PRESENT as classically defined (structural analog noted).
   Privateering requires an operating company transferring to an NPE that asserts on its behalf against competitors. Yonsei's UIF is a non‑profit university tech‑transfer office, not a competitor of codec implementers. The analog worth flagging: university SEP monetization via an NPE intermediary, where the university appears to have exited completely at reel 056358/0816. Whether Yonsei retained a royalty interest or license‑back is not determinable from the recorded assignment abstract.

8. Defensive aggregator — NOT PRESENT (inverse situation).
   The chain does not terminate at RPX, AST, LOT, Unified or OIN. The opposite is true: Unified Patents — a defensive aggregator — is attacking this patent (ex parte reexam 90/019,837, filed 2025‑01‑31), which confirms the patent is being asserted rather than neutralized.

Verdict

NPE — high confidence

The patent left a non‑practicing university foundation in reel 056358/0816, effective 2021‑05‑14 (recorded 2021‑05‑26), landing at Ideahub Inc. — an entity expressly characterized as an NPE by Unified Patents, with no product embodying the claims, which monetizes its portfolio through an exclusive‑license litigation vehicle (Helios Streaming, LLC) and through the Access Advance VVC pool, where the '539 is declared essential (Signal 1 + Signal 2, both strong). Corroborating conduct: Ideahub has been a repeat plaintiff on other patents in the same program (Vudu, Crackle, Showtime, Starz), and a defensive aggregator paid to attack this patent via ex parte reexamination in 2025 — the reexam certificate of 2026‑02‑03 confirms Ideahub still owns it and that claims 1‑5 survived with claims 6‑20 added.

Two verification gaps to close before relying on this in a brief: (a) the correspondent of record on reels 038879/0223 and 056358/0816, and (b) whether the corresponding Ariscale → Ideahub recordings share that correspondent (Signal 3). Verify at: USPTO Assignment Center or assignment.uspto.gov — search patent number 10904539 / reel 056358 frame 0816, and cross‑check the ideahub.co.kr contact of record, Kyeongsu Im.

I'll verify the citation data against the USPTO/Google Patents record and gather details on each cited reference.

Let me gather details on the remaining cited references.

Prior Art Analysis — U.S. Patent No. 10,904,539 B2

Patent: US 10,904,539 B2 — "Method, device and system for encoding and decoding image"
Application: US 15/026,154 (U.S. national stage of PCT/KR2014/009221); PCT filed 2014-09-30; priority KR 10-2013-0116220 (2013-09-30)
Effective filing date: 2013-09-30 (AIA applies — the PCT was filed after 2013-03-16, so AIA §§ 102(a)(1)/(a)(2) govern)
Assignee: Ideahub Inc. (orig. University-Industry Foundation, Yonsei University)

Data-source note / caveat. The citation set below is taken from the authoritative Google Patents record for US10904539B2 (https://patents.google.com/patent/US10904539/en), which lists 13 patent citations and 6 non-patent citations. My direct USPTO-PatentCenter retrieval was not available through my search tools, so I worked from that record plus corroborating searches. Where I could not retrieve a reference's specification text, I say so explicitly rather than inferring its content. I did not substitute any similar patent number (e.g., 10,949,339 is a different Netlist memory patent and is excluded).

Structural caveat from the prior sections (carried forward, not repeated): the ’539 patent is not a substitute for the record — claims 1–5 were confirmed in the Unified Patents ex parte reexamination (Reexamination Certificate C1, decision 2026-02-03), and new claims 6–20 were added (text not available). That confirmation is itself evidence that the CRU did not treat any cited reference as anticipatory.

A. The 13 cited patent references

The record flags a subset as examiner-cited (asterisk in Google Patents) versus third-party-cited, but the collapsed table formatting makes the exact split unreliable; treat the categorization as approximate.

Per-reference description and § 102 assessment (mapping to the claim elements: [A] LM-mode flag signaling; [B] index/mode signaling of the selected candidate; [C] two LMmode candidates, at least one using a filtered luma value; [D] weight/offset derived from a "second filtered luma value" only under the first LMmode; [E] first filtered luma value = weighted sum of co-located + horizontal + vertical neighbor; [F] Cpred = α·G_recon_lpf[x,y] + β):

1–5, 7, 9, 13 (general intra-prediction / unrelated art) — no anticipation of any claim.

• US 7,227,585 B1 (Conexant): luma/chroma separation of a composite video signal. Discloses filtering to separate luma from chroma — background only; nothing on [A]–[F]. No anticipation.
• JP 4989048 B2 (Microsoft): layered/embedded 3D-subband video coding. Unrelated to LM chroma modes. No anticipation.
• US 2007/0115384 A1 (Olympus): distortion-aberration correction. No anticipation.
• JP 2009-538086 A (LG): generic video-signal coding/decoding. No LM-mode teaching. No anticipation.
• KR 10-2012-0041287 A (SK Telecom): intra-prediction codec, but no two-LM-candidate structure, no filtered luma, no flag+index LM signaling. No anticipation.
• KR 10-2013-0004214 A (Sungkyunkwan): multi-level prediction; unrelated to LM. No anticipation.
• KR 10-2013-0078320 A (GIST): depth-image coding. No anticipation.
• KR 10-2013-0044263 A (SK Telecom): HD video codec; general framing only. No anticipation.

6. US 2013/0182761 A1 / US 9,661,337 B2 — Samsung Electronics (Jianle Chen, Vadim Seregin, et al.) — Tier-1 LM art.
Description: obtains parameters representing the correlation between a chroma block and its corresponding luma block from restored neighboring luma/chroma pixels; down-samples luma; predicts the chroma block from luma using those parameters, with adaptive scaling to avoid overflow. This is the patent-family counterpart to the Chen CE6.a.4 work (see NPL below).
§ 102: Discloses the base LM prediction — i.e., chroma predicted as a linear function of reconstructed luma with weights/offset derived from neighboring luma and chroma ([F], and the "linear relationship between luma pixels and chroma pixels" of claim 3/5). It does not disclose two selectable LMmode candidates ([C]), a filtered-luma LM variant, the [D] weight/offset distinction, the [E] three-neighbor weighted sum, or the flag+index signaling ([A],[B]).
Verdict: does not anticipate claims 1, 3, 4, or 5 in full; strong § 103 reference.

8. US 2013/0336591 A1 / US 10,009,631 B2 — LG Electronics (Jeon, Park, Lim, Kim, Park, Choi, Sung, Jeon) — Tier-1 signaling art.
Description: derives the intra chroma prediction mode using an LM mapping table when LM is used; uses different codeword mapping tables depending on chroma_pred_from_luma_enabled_flag (e.g., value 4 = LM, value 5 = DM; the LM/DM bits are context-coded, remaining bits bypass-coded). Note: this is the closest cited reference on the signaling limitations.
§ 102: Discloses [A] a flag indicating whether luma-based ("LM") chroma prediction is used and [B] codeword/mode-index information specifying the chroma prediction mode — arguably an "index specifying one of the candidates." But it discloses only one LM mode (no filtered-luma LM candidate), so [C], [D], [E] are absent.
Verdict: does not anticipate claims 1/3/4/5 in full; the strongest § 103 reference for the flag+index element. Prior-art status rests on § 102(a)(2) (effective filing 2011-03-05/2012-03-05) because its 2013-12-18 publication postdates the ’539 effective date.
Dating footnote (literal): the ’539 record lists this reference's priority as 2011-03-06; the Unified Patents portal lists 2011-03-05. I report both rather than choosing.

10. US 9,307,237 B2 — Futurewei Technologies (Liu, Li, Ling, Zheng, Zhang, Song) — Tier-1 filtered-luma art.
Description: generates a chroma prediction block whose predicted chroma sample is based on a filtered reconstructed luma sample in the corresponding reconstructed luma block, plus downsampled filtered reconstructed luma samples in neighboring blocks and downsampled neighboring chroma samples; when the chroma block exceeds a threshold, it downsamples the filtered luma and chroma neighbors (i.e., averages co-located and adjacent samples).
§ 102: This is the closest cited reference to [E] (a filtered luma value derived by combining a co-located luma sample with neighboring luma samples) and to "[a] first filtered luma value" generally (support also for [C]'s "filtered luma"). It does not disclose a pair of LMmode candidates distinguished by whether the weight/offset derivation uses a filtered luma value ([C],[D]), nor the flag+index LM-candidate signaling ([A],[B]). (Same-family sibling US 9,438,904 B2 exists but is not in the ’539 citation list.)
Verdict: does not anticipate claims 1/3/4/5 in full; best § 103 combination partner with the LM-signaling references. Prior art via § 102(a)(2) (effective 2012-01-19/2013-01-18).

11. US 9,351,001 B2 — Huawei Technologies — relevance unconfirmed.
Description: "'Encoding or decoding method and apparatus" — filed 2012-01-20, granted 2016-05-24. I could not retrieve this reference's specification text within my search budget, so I cannot map its disclosure to [A]–[F]. Based on its citation context (listed alongside the LM-mode art), it appears directed to intra/luma-based chroma coding, but I will not assert that.
Verdict: cannot be confirmed as anticipatory on the available record; flagged as uncertain — this reference should be pulled and read directly before any § 102 reliance.

12. US 2015/0078447 A1 — Sony (Silcock, Sharman, Saunders, Gamei) — chroma-mode-derivation art.
Description: video coding/decoding in which luma and chroma samples are predicted from other reference samples per a prediction direction, with direction mapping between grids of differing aspect ratios where chroma has a lower sampling rate (§102 counterpart: GB priority 2012-04-26; US filed 2013-04-25).
§ 102: directed to chroma intra prediction-mode derivation and direction remapping, not to LM filtering or two-LM-candidate selection. Discloses no [A]–[F] combination.
Verdict: does not anticipate claims 1/3/4/5; general chroma-intra-prediction background. Prior art via § 102(a)(2) (effective 2013-04-25).

B. The 6 cited non-patent references

N1. "Development of HEVC Based Coding Tools for Efficient Screen Content Coding," Nov. 2012.
Description: HEVC-based coding tools for screen content. § 102: Nothing in the record shows it discloses the two-LM/flag+index/filtered-luma combination. No anticipation; background.

N2. B. K. Gunturk, Y. Altunbasak, R. M. Mersereau, "Color plane interpolation using alternating projections," IEEE Trans. Image Processing, vol. 11, no. 9, pp. 997–1013, Sept. 2002.
Description: color-plane interpolation via alternating projections; the source of the LL/LH/HL/HH inter-plane correlation analysis reproduced in the patent's Table 1. § 102: discloses inter-plane correlation analysis and filtering/interpolation between color planes, but not video-coding LM modes or their signaling. No anticipation of claims 1/3/4/5; supports the § 103 rationale for low-pass filtering high-frequency content where plane correlation is low.

N3. J. Chen, V. Seregin, W-J Han, J. Kim, B. Jeon, "CE6.a.4: Chroma intra prediction by reconstructed luma samples," JCTVC-E266, 5th JCT-VC Meeting, Geneva, Switzerland, March 2011. (cited twice by the examiner)
Description: the foundational LM-mode contribution — predicts chroma from reconstructed luma via a linear model Cpred = α·recL + β, with α and β inferred from neighboring reconstructed luma/chroma samples.
§ 102: the closest single reference to the base prediction recited in the claims ([F]; and claim 3/5's "linear relationship between luma pixels and chroma pixels"). The patent itself concedes Chen as prior art. It discloses only one LM mode, no filtered-luma variant, and no flag/index candidate selection ([C],[D],[E],[A],[B] absent).
Verdict: does not anticipate claims 1/3/4/5 in full; strongest § 103 base reference.

N4. J. Kim, "RCE1: The performance of extended chroma mode for non 4:2:0 format," JCTVC-M0097, 13th JCT-VC Meeting, Incheon, Korea, April 2013. (cited twice by the examiner)
Description: the extended chroma mode (ECM) for non-4:2:0/RGB formats — chroma predicted from reconstructed luma. The ’539 specification expressly attributes the "LM chroma mode" (its second LMmode) to J. Kim.
§ 102: discloses the second LMmode of the claims (luma→chroma prediction for RGB 4:4:4), but no low-pass-filtered variant, no second candidate, and no flag+index signaling ([C],[D],[E],[A],[B] absent).
Verdict: does not anticipate claims 1/3/4/5 in full; § 103 reference for the "LM chroma mode" element. Prior art under § 102(a)(1) (April 2013 printed publication, before the 2013-09-30 effective date).

N5 / N6. Duplicate record entries of Chen CE6.a.4 and Kim RCE1 as examiner-cited.

C. § 102 bottom line

No single cited reference discloses every limitation of claims 1, 3, 4, or 5. The three limitations that defeat § 102 across the entire citation set are:

1. [C]/[D] — a plurality of LMmode candidates (a first, filtered-luma LMmode and a second LMmode) where the weight/offset derivation differs by whether a "second filtered luma value" is used. Neither Chen (N3), Kim (N4), Samsung (6), LG (8), nor Futurewei (10) shows two LM variants distinguished at the parameter-derivation stage.
2. [A]/[B] — encoding/parsing a flag for LMmode application plus index information identifying the selected LMmode candidate. LG (8) is closest on flag+index but for LM-vs-non-LM, not for choosing among multiple LM candidates.
3. [E] — the first filtered luma value computed as a weighted sum of the co-located luma pixel together with a horizontal and a vertical neighboring luma pixel. Futurewei (10) is closest via its filtered/downsampled reconstructed luma samples, but it does not present this as the claim-1 weighted-sum construction.

Best element-level (§ 102-in-part / § 103) mapping:

Practical § 103 combinations a challenger would run: (i) Chen CE6.a.4 or Kim RCE1 (LM/ECM, [F]) + Futurewei US9307237B2 (filtered luma, [C]/[E]) + LG US20130336591A1 (flag/index signaling, [A]/[B]); with Gunturk (N2) supplying the motivation to low-pass filter high-frequency, low-correlation color-plane content. That the CRU nonetheless confirmed claims 1–5 (Reexamination Certificate C1) is consistent with this analysis — no cited reference was treated as anticipatory.

Claim 2: dependent on claim 1 (left/right neighbor for the horizontal pixel; top/bottom for the vertical). Since no reference anticipates claim 1, claim 2 cannot be anticipated; it would fall or stand with claim 1.

D. Items that are not § 102 prior art (flagged to avoid misuse)

• Forward citations (documents citing the ’539 family) — US 2023/0055497 A1 (Hyundai, 2020 priority) and the "Families Citing this Family" list (CN 103220508 B, KR 101530774 B1, KR 101530782 B1, CN 109274969 B, CN 115834897 B, PE 20241804 A1, WO 2021004152 A1, KR 102762371 B1) — are later documents and cannot be § 102 art against the ’539 priority date.
• Family members (US 11,252,421; US 11,683,506; US 12,301,842; US 2025/0267293 A1; KR 101530774 B1; WO 2015/047039 A1) share the ’539 priority and are not separate prior art.

E. Confidence and gaps

• High confidence on the citation list, dates, and claim-element mapping, all drawn from the authoritative Google Patents record (https://patents.google.com/patent/US10904539/en) and corroborated for refs 6, 8, 10, 12 and N3/N4.
• Explicitly uncertain: (a) the full specification text of US 9,351,001 B2 (Huawei) — I could not retrieve it, so its § 102 relevance is unverified; (b) the exact examiner-vs-third-party split of the 13 citations; (c) the text of new claims 6–20 from the reexamination, which could not be assessed and which may cover different scope.
• Recommended next step: pull the native USPTO file wrapper and the C1 reexamination certificate for application 15/026,154 / reexam control 90/019,837 (per the prior section) to (i) confirm the citation categorization and (ii) obtain claims 6–20 for a complete § 102 analysis.

I'll ground the reference mappings with a few targeted searches before analyzing, since the prior-art record gives me identifiers and titles but not full reference text.

Obviousness Analysis — U.S. Patent No. 10,904,539 B2 under 35 U.S.C. § 103

Analysis date: April 26, 2026 (note: the previously generated sections in this thread are dated 2026-09-29; I am proceeding on the date given in this task. The substantive record is unaffected.)
Patent analyzed: US 10,904,539 B2, Method, device and system for encoding and decoding image (Ideahub Inc.; priority 2013-09-30; issued 2021-01-26). Claims analyzed: issued claims 1–5 only. I still do not have the text of reexam-added claims 6–20 and therefore express no opinion on their patentability.

1. Governing framework and the prior-art corpus

Law applied. The '539 has an effective filing date of 2013-09-30 (PCT/KR2014/009221, filed 2014-09-30). That is after 2013-03-16, so the AIA version of § 102/§ 103 governs. Obviousness is assessed under KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007), and the Graham v. John Deere factors, using the MPEP 2143 rationales.

Prior-art corpus used. Per instruction, I am working from the references listed in the Prior Art sections of the patent page:

Examiner/third-party Citations (13): US 7,227,585 B1 (Conexant); US 2007/0115384 A1 (Olympus); JP 4989048 B2 (Microsoft); JP 2009-538086 A (LG); KR 10-2012-0041287 A (SK Telecom); US 2013/0182761 A1 (Samsung); KR 10-2013-0004214 A; US 2013/0336591 A1 (LG Electronics — "Intra prediction method of chrominance block using luminance sample"); KR 10-2013-0078320 A; US 9,307,237 B2 (Futurewei — "Reference pixel reduction for intra LM prediction"); US 9,351,001 B2 (Huawei); US 2015/0078447 A1 (Sony — "Intra prediction mode derivation for chrominance values"); KR 10-2013-0044263 A (SK Telecom).

Non-Patent Citations (6): J. Chen et al., "CE6.a.4: Chroma intra prediction by reconstructed luma samples," JCTVC-E266 (Mar. 2011); J. Kim, "RCE1: The performance of extended chroma mode for non 4:2:0 format," JCTVC-M0097 (Apr. 2013); B.K. Gunturk et al., "Color plane interpolation using alternating projections," IEEE Trans. Image Processing 11(9):997–1013 (Sep. 2002); and "Development of HEVC Based Coding Tools for Efficient Screen Content Coding" (Nov. 2012).

Family Cites Families (20) — cited against sibling family members, so not necessarily before the '539's examiner: US 9,288,500 B2 (Texas Instruments — "Luma-based chroma intra-prediction for video coding"); US 9,693,070 B2 (TI); CN 103782596 A (Samsung); US 9,948,938 B2 (TI); GB 2 495 941 B (Canon); GB 2 498 550 B (Canon); WO 2013/189205 A1 (MediaTek).

⚠️ Disclosure of method: I could not retrieve the full text of every listed reference. Where I characterize a reference's teaching, I relied on its abstract, claims, or its direct family members (e.g., the LG publication's granted counterparts US 9,800,894 / US 10,009,631; the Canon GB's US equivalent US 9,615,102; the TI '500 continuation US 11,197,024). That substitution is sound for the family members, but a formal invalidity chart must be built on the specific cited document texts.

Level of ordinary skill (PHOSITA). A person with a B.S. in EE/CS (or equivalent) and 2–4 years of video-compression experience, including working familiarity with the HEVC draft standard, the HM reference software, the HEVC Range Extensions activity, and JCT-VC contributions — i.e., someone who would routinely read JCTVC-E266, JCTVC-M0097, and the Access Advance/VVC-era LM/CCLM literature.

2. The inventive delta is narrow — and the specification says so

This matters for the obviousness posture. The '539's own Background admits the ECM/LM chroma mode as prior art:

"an extended chroma mode … has been suggested by J. Kim at a recent MPEG standardization conference (J. Kim, 'RCE1: The performance of extended chroma mode for non 4:2:0 format,' JCTVC-M0097 …April 2013)"

And the Summary states the point of novelty:

"the intra predicting unit 120 provides not only chroma modes provided by a conventional intra prediction unit, which include a planar mode, a vertical mode, a horizontal mode, and a DC mode, but also an LM chroma mode, which is an ECM suggested by J. Kim, and a low pass filter (LPF) LM chroma mode, which is a mode for applying an ECM after removing high frequency ingredients of a Luma (G) area by using a LPF according to the present invention."

Strip away the admitted prior art and what remains is: (1) low-pass filter the luma before applying LM prediction; (2) offer both the filtered and unfiltered variants as competing candidates; (3) signal which one was used. Each of those three is independently known; the issue is only whether their combination was obvious. It plainly was.

Claim-construction caution (governs everything below). Claim 1 introduces a "first filtered luma value" and a "second filtered luma value" without defining either, and recites that α and β are "obtained based on a second filtered luma value" under the first LMmode but "in absence of the second filtered luma value" under the second LMmode. Both the prediction equation and the claim body use inconsistent symbols (G lpf recon vs. G_recon_lpf). The most coherent reading — and the one that tracks the specification — is:

• "first filtered luma value" = the spatially filtered co-located luma sample used in C_pred[x,y] = α·G_lpf_recon[x,y] + β;
• "second filtered luma value" = the spatially filtered neighboring luma samples (top/left template) from which the linear-model parameters α and β are derived;
• first LMmode = filtering applied; second LMmode = no such filtering.

I flag this because a narrower construction of "filtered" could matter, and because the same ambiguity is an independent § 112 problem (see § 9).

3. GROUND 1 (primary): Chen (JCTVC-E266) + Kim (JCTVC-M0097) + Gunturk + LG '591

3.1 The references

3.2 The LG reference is devastating on the signaling limitations

LG '591's codeword mapping table (retrieved from the published application) is reproduced below, verbatim in structure:

Sources: https://www.freepatentsonline.com/y2013/0336591.html ; https://patents.google.com/patent/US20130336591A1/en ; https://portal.unifiedpatents.com/patents/patent/US-20130336591-A1

Now compare the '539's own Table 2:

The '539's flag is literally the same identifier — chroma_pred_from_luma_enabled_flag — that LG '591 discloses, and its prefix scheme ("0" for one mode, "10" for another, "11"+suffix for the conventional modes) is LG's scheme with one additional candidate inserted into the reserved "10" branch. This is not a coincidental resemblance; it means the "flag" and "index information" limitations of claims 1, 3, 4 and 5 were old in the art before the '539's priority date. The '539's contribution on the syntax side is a two-bit suffix — a routine binarization choice driven by the need to encode one more mode.

3.3 Why a PHOSITA would have made the combination

1. Same field, same workstream, same authors' dialogue. E266 and M0097 are both JCT-VC inputs to the same HEVC Range Extension process; the '539 itself states that Kim's ECM "employs a technique suggested by J. Chen." Combining two contributions to the same standardization effort is the paradigm KSR combination.
2. Gunturk supplies the articulated problem. Gunturk's analysis (and the '539's Table 1, reproduced from it) shows inter-plane correlation of ≈0.90 in the LL band but ≈0.38 in the HH band. Once a PHOSITA knows that the high-frequency luma content is precisely the content that does not correlate with chroma, the advantage of attenuating that content before using luma as a chroma predictor is not merely suggested — it is the analytically identified solution. KSR ("finite number of identified, predictable solutions") and MPEP 2144.04 apply directly.
3. Low-pass filtering is a known technique with a known, predictable effect — MPEP 2143(A)(3) ("use of a known technique to improve a similar device in the same way"). Applying a 2-D [1 2 1]/4 smoothing kernel is elementary.
4. Filtering luma before cross-component prediction was already done. E266's LM mode already filters/subsamples the reconstructed luma (the [1 2 1]/4 kernel) before using it to predict chroma. The '539 does not claim downsampling; it claims a weighted-sum filtered luma value. Changing why the same filter is applied (de-correlation rather than resolution matching) is a change in the intended purpose, not in the physical steps.
5. Two-candidate competition with RDO selection is the standard HEVC pattern, not an inventive step: E266/LG '591 have LM competing against planar/vertical/horizontal/DC; the '539's own FIG. 4 ("generate prediction blocks in all available modes … rate-distortion optimizing unit … determine an optimal mode") describes the ordinary encoder search.

3.4 Element-by-element chart — Claim 1

Result: every element of claim 1 is disclosed or rendered obvious. The only genuine gap — a filtered parameter-derivation branch paired with an unfiltered branch — is closed by US 9,288,500 (Ground 2), and the motivation for filtering comes from Gunturk. Ground 1 is a strong, defensible § 103 case; Ground 2 closes the last gap.

4. GROUND 2 (belt-and-suspenders): TI US 9,288,500 + LG '591 + Gunturk

US 9,288,500 B2 (priority 2011-05-11; filed 2012-05-10; § 102(a)(2)/(e) art) discloses a luma-based chroma intra-prediction method that:

"includes filtering reconstructed neighboring samples of a reconstructed down sampled luma block, computing parameters α and β of a linear model using the filtered, reconstructed neighboring samples … wherein the linear model is Pred_C[x,y] = α·Rec_L′[x,y] + β … and computing samples of a predicted chroma block from corresponding samples of the reconstructed down sampled luma block using the linear model and the parameters."
(https://portal.unifiedpatents.com/patents/patent/US-[9288500](/patent/9288500)-B2)

That is claim 1's "second filtered luma value" limitation, literally: the weight and offset are derived from filtered luma samples taken from the neighboring region. The corresponding TI continuation describes the full two-stage geometry expressly:

"The interior pixels of the reconstructed luma block 1100 are filtered and down sampled to generate a filtered, down sampled luma block 1102 for generating the predicted chroma block. The reconstructed top neighboring pixels and the reconstructed left neighboring pixels … are also filtered … for use in deriving the parameters α and β."
(https://www.freepatentsonline.com/[11197024](/patent/11197024).html)

That is: a filtered co-located/interior luma value feeding the prediction equation, and a separate filtered neighboring luma value feeding the α/β derivation — i.e., the "first" and "second" filtered luma values of claim 1. TI also discloses plural LM intra-prediction types signaled from encoder to decoder ("If the type is 0 … If the type is 1 … If the type is 2 … In the decoder, the type used by the encoder is decoded from the encoded bit stream"), which is the "plurality of LMmode candidates"+index concept. TI further discloses that for 4:4:4 sampling the luma is not downsampled — but it does not exclude filtering, and in any event the '539 does not claim downsampling.

Combination: TI '500 (filtered-luma LM with filtered parameter derivation; plural signaled LM types) + LG '591 (LM mode; flag naming and codeword index for selection) + Gunturk (the reason to filter) + Kim M0097 (RGB 4:4:4 application context). Motivation: (i) TI '500 opens with the same problem (compression efficiency of luma-based chroma prediction) in the same field; (ii) Gunturk identifies the frequency-band problem and thus the solution direction; (iii) multiplexing a filtered and an unfiltered variant and letting RDO choose is the ordinary encoder-selection practice both TI '500 and LG '591 already employ; (iv) the filtering of a signal to attenuate a known low-correlation component is a predictable use of a known technique.

Why this matters strategically. TI '500 appears in the record only in the "Family Cites Families" list — i.e., it was cited in a sibling's file, not on this patent's face. It therefore was likely not considered by the '539's examiner or by the reexam examiner. That makes it the natural lead reference for a § 103 challenge, and it sidesteps the § 325(d) "same art previously presented" risk that the previously generated sections correctly flagged.

5. GROUND 3 (secondary/alternatives)

• Futurewei US 9,307,237 B2 ("Reference pixel reduction for intra LM prediction," priority 2012-01-19) — modifies which and how many reference luma pixels feed the LM parameter derivation. It reinforces that the LM reference-pixel set and its preparation were a recognized design variable, and it teaches substituting a differently-filtered/reduced luma reference set in LM prediction — MPEP 2143(A)(2) (substitution of one known element for another).
• Canon GB 2 495 941 B / US 9,615,102 B2 — intra coding of chroma samples predictable from luma samples "using a prediction model," with per-block parameter computation and adaptivity (parameter error detection/correction) (https://patents.google.com/patent/GB2495941B/en). Useful as corroboration of the breadth of the inter-component-prediction design space and of per-block adaptation, not as the primary teaching.
• Sony US 2015/0078447 A1 ("Intra prediction mode derivation for chrominance values") and MediaTek WO 2013/189205 A1 ("adaptive intra prediction") — corroborate that deriving and signaling a chroma prediction mode (including from luma) was routine, supporting the flag/index limitations.
• Samsung US 2013/0182761 A1 and SK Telecom KR 10-2012-0041287 A — intra prediction with reference-pixel filtering (e.g., mode-dependent intra smoothing) and encoders that evaluate multiple intra modes. These support the proposition that filtering samples before prediction and RDO-based mode selection were conventional.

6. Claims 2–5

• Claim 2 (H-neighbor = left/right; V-neighbor = top/bottom). Adds nothing: E266's 2-D [1 2 1]/4 luma kernel uses the left and right neighbors horizontally and the top/bottom neighbors vertically; TI '500 uses [0.25, 0.5, 0.25]-type filters with distinct top and left neighbor filters. The dependent claim is squarely within the same combination.
• Claims 3 and 5 (decoder/dedecoding method). LG '591 is expressly directed to a decoding method — "calculating an intra prediction mode of a chrominance block on the basis of an LM mapping table," "decoding information on whether the LM is used," "receiving codeword information," and deriving α/β by the expressions it sets out. TI '500 expressly covers "a video encoder or a video decoder," and its continuation states the LM type is "decoded from the encoded bit stream." Claims 3/5 add the further limitation that the weight and offset are obtained "based on a linear relationship between luma pixels and chroma pixels," which is the definition of LM and is disclosed by E266, Kim, LG '591 and TI '500 alike.
• Claim 4 (encoding method) is the method counterpart of claim 1 and is met by the same references.

7. KSR rationales, expressly stated

For the record, the enumerated rationales are:

1. Combining prior-art elements to yield the claimed invention — MPEP 2143(A)(1): E266/Kim (LM prediction for RGB 4:4:4) + Gunturk (low HF correlation) + a low-pass filter.
2. Simple substitution of one known element for another — MPEP 2143(A)(2): using the already-known [1 2 1]/4 luma filter in a de-correlation role rather than a downsampling role.
3. Use of a known technique to improve a similar device in the same way — MPEP 2143(A)(3): low-pass smoothing, and RDO-based candidate selection, applied to the LM prediction pipeline.
4. Applying a known technique to a known device ready for improvement — MPEP 2143(A)(4): the LM/ECM predictor was explicitly under improvement for the HEVC Range Extension.
5. Obvious to try / finite number of identified, predictable solutions — MPEP 2144.04: filtering the luma (or not) is one binary design choice at a single point in a known pipeline, and the spatial distribution maps in the '539's own FIG. 1B/1C make clear that neither choice dominates everywhere.
6. Predictable results — the '539's own experiments (its Table 3, reproduced from the inventors' July 2013 Korean-language paper at http://koreascience.kr/article/JAKO201322658550184.pdf) show BD-rate changes of only −0.4% / −0.6% / −0.8% for G/B/R, i.e., ordinary filter-tuning magnitudes.

8. What the patent owner will argue, and why it likely fails

(a) "TI '500 teaches away from filtering in 4:4:4." TI's 4:4:4 passage says the luma block and neighboring luma samples "are not down sampled" where luma/chroma sampling rates match. But (i) the '539 claims filtered luma, not downsampled luma; (ii) TI's filtering of the neighboring samples is recited in TI's claim 1 independently of downsampling; and (iii) the '539 supplies no evidence that a PHOSITA would read "no downsampling" as "no filtering." A teaching-away argument requires a reference that "criticizes, discredits, or otherwise discourages" the claimed approach — In re Fulton; that standard is not met here.

(b) "The references do not disclose two LM candidates distinguished by filtering." True of any single reference — which is precisely why the law permits combination. The gap is a binary option at one pipeline node, and the reason to keep both is supplied by the spatial variability the '539's own FIG. 1B/1C demonstrate (high- and low-correlation regions coexisting in a single frame). Offering an encoder two alternatives and letting rate-distortion optimization choose is the ordinary, well-documented HEVC encoder architecture (LG '591; TI '500; '539 FIG. 4).

(c) "Unexpected results / secondary considerations." The record cuts the other way:

• The measured benefit is small and not uniform. The '539's own Table 3 shows EBURainFruits G = +0.6% — a loss. The inventors' own Korean paper explains this: for that sequence the planes correlate highly across all frequency bands, so filtering helps nothing "and rather, the additional mode-encoding bits cause a BD-rate loss." A result that is predictable from the known frequency-band analysis is the opposite of an unexpected result; it is confirmation that the invention is the straightforward application of Gunturk's teaching.
• The '539 discloses no comparative data against the closest prior art that would isolate the alleged invention's contribution; Table 3 compares only to the conventional ECM.
• Encoder complexity rose 8% (Dec Time 101%) — consistent with adding one more candidate to an existing search, not with a surprising technical effect.

(d) "The specific 0.5/0.125×4 kernel is inventive." Not claimed. Claim 1 requires only "a weighted sum operation of the co-located luma pixel, the horizontal neighboring luma pixel and the vertical neighboring luma pixel"; claim 2 only names the left/right/top/bottom neighbors. The specific Equation 3 weights are unclaimed subject matter and cannot rescue the claims.

Net assessment:

Because claims 1–5 were expressly confirmed patentable in the reexamination (decision dated 2026-02-03, kind code C1, effective 2026-01-26), a challenger should expect the patent owner to argue that the reexam examiner already passed on the LM art — which makes it important that the lead references be art the Office likely did not consider (see § 9).

9. Practical and procedural notes

1. Lead with art outside the face of the patent. LG '591 was before the examiner (it is one of the 13 citations), so it is a § 325(d) magnet at the PTAB. TI US 9,288,500 (and its relatives US 9,693,070 / US 9,948,938) appears only in the Family Cites Families list and is the better lead reference, with Gunturk (cited, but only for its correlation analysis, not for filtering) supplying the motivation. Futurewei US 9,307,237 and the Canon GBs are additional non-face art.
2. § 112 is arguably the sharper sword. The demonstrated vulnerability in this family has been written description (Unified's wins on US 9,641,849 and US 11,122,274, affirmed on appeal). The '539's claim language has an evident antecedent-basis/consistency problem: "first filtered luma value" vs. "second filtered luma value" appear without definition; "obtained in absence of the second filtered luma value" is of uncertain scope; and the equation symbol G lpf recon does not match the claim body's G_recon_lpf. Even so, the reexam examiner confirmed these claims, so this must be developed with the reexam file history in hand. (This is a flag, not a legal opinion on indefiniteness.)
3. Investigate the inventors' July 2013 publication. A Korean-language journal paper at http://koreascience.kr/article/JAKO201322658550184.pdf — JBE (Journal of Broadcast Engineering) Vol. 18, No. 4, July 2013 — reproduces Tables 2 and 3 identically to the '539. It is dated roughly two months before the 2013-09-30 priority date. If its authorship is exactly the named inventors (Yoon Sik Choe; Jang Won Choi), it falls within the § 102(b)(1)(A) grace period and is not prior art. If a non-inventor contributed to it, it is a § 102(b)(1)(A)/(B) question and potentially a § 102(a)(1) printed publication predating the priority date. This should be the first document pulled. I have not verified its author list and state no conclusion.
4. § 102 (anticipation) is not the right vehicle for this patent; the claims' two-candidate architecture and the "second filtered luma value" limitations are not found in a single reference. § 103 is where the exposure is.
5. Scope limit. This analysis covers only issued claims 1–5. Claims 6–20, added by the reexamination certificate, have not been obtained or analyzed, and no opinion is offered on them.

10. Bottom line

On the record available, issued claims 1–5 of US 10,904,539 are highly vulnerable under § 103. The reference set on the patent's own face supplies: the LM/chroma-from-luma predictor (Chen, JCTVC-E266), its extension to RGB 4:4:4 (Kim, JCTVC-M0097), the exact flag-and-index syntax for selecting and signaling it (chroma_pred_from_luma_enabled_flag and the codeword table of LG US 2013/0336591 — the same flag name the '539 uses in its own Table 2), and the analytical basis for the only asserted improvement (Gunturk: inter-plane correlation is high in LL but low in HH, i.e., precisely where filtering would be applied). US 9,288,500 (Texas Instruments) — cited only against the family, not against this patent — supplies the remaining element, teaching a linear LM model Pred_C = α·Rec_L′ + β in which both the prediction reference and the neighboring samples used to derive α and β are filtered, and in which plural LM types are signaled in the bitstream. Nothing beyond routine mode competition and rate-distortion selection is left. The patent's own measured benefit (0.4–0.8% BD-rate, with a documented 0.6% loss on one sequence) is small, non-uniform and predictable from the very correlation analysis the specification borrows from Gunturk — which defeats, rather than supports, a secondary-considerations case.

Confidence: High on the claim-element mappings drawn from the retrieved reference texts (LG '591, TI '500/its continuations, Canon/US 9,615,102, and the '539's own specification). Moderate-to-high on the overall conclusion, with the caveats that (i) I do not have the full text of every one of the 13 examiner-cited references, (ii) the reexam record and the text of new claims 6–20 are unavailable to me, and (iii) the claim-construction ambiguity flagged in § 2 means a court could narrow "filtered luma value" in a way that complicates part of the chart.