DCT
1:23-cv-00136
Nielsen Co US LLC v. Hyphametrics Inc
I. Executive Summary and Procedural Information
- Parties & Counsel:
- Plaintiff: The Nielsen Company (US), LLC (Delaware)
- Defendant: HyphaMetrics, Inc. (Delaware)
- Plaintiff’s Counsel: Potter Anderson & Corroon LLP
- Case Identification: 1:23-cv-00136, D. Del., 02/03/2023
- Venue Allegations: Venue is alleged to be proper in the District of Delaware because the Defendant is a Delaware corporation.
- Core Dispute: Plaintiff alleges that Defendant’s audience measurement products and methods infringe a patent related to using deep neural networks to automatically detect graphical overlays in images.
- Technical Context: The technology operates within the field of automated content recognition (ACR), which is critical for media companies and advertisers to measure audience engagement and verify ad placement across broadcast and streaming platforms.
- Key Procedural History: The patent-in-suit claims priority from two provisional applications filed in November 2016. The complaint relies significantly on a patent owned by the Defendant (U.S. Patent No. 10,932,002) to describe the functionality of the accused products.
Case Timeline
| Date | Event |
|---|---|
| 2016-11-07 | '588 Patent Earliest Priority Date |
| 2021-03-25 | "Cohen Article" describing HyphaMetrics published |
| 2021-04-06 | '588 Patent Issued |
| 2021-10-13 | HyphaMetrics "VideoAmp Deal" announced |
| 2023-02-03 | Complaint Filed |
II. Technology and Patent(s)-in-Suit Analysis
U.S. Patent No. 10,970,588 - Recurrent Deep Neural Network System for Detecting Overlays in Images
The Invention Explained
- Problem Addressed: The patent addresses the inefficiency of conventional techniques for identifying and removing graphical overlays (e.g., channel logos, news tickers) from images and video streams, which often require manual human review to mark the unwanted regions. (U.S. Patent No. 10,970,588, col. 3:3-8).
- The Patented Solution: The invention proposes an automated, multi-stage method using neural networks. First, a "feature map network" processes an image to create a data representation, or "feature map," that characterizes graphical objects within it. A second "probability map network" then analyzes this feature map to generate a probability score for different regions of the image, indicating the likelihood that they contain an overlay. If a region's score exceeds a set threshold, a subsequent action is triggered, such as channel identification or ad-compliance tracking. ('588 Patent, Abstract; col. 6:39-67).
- Technical Importance: This automated approach is described as an improvement to the technical field of image processing by increasing the speed and reliability of overlay detection while reducing or eliminating the need for human oversight. ('588 Patent, col. 3:18-22).
Key Claims at a Glance
- The complaint asserts independent claims 1 (a method) and 10 (a non-transitory computer-readable medium). (Compl. ¶41).
- Independent Claim 1 requires:
- A processor applying a feature map network to an image to create a feature map, which comprises a grid of vectors characterizing a feature.
- The processor applying a probability map network to the feature map to create a probability map assigning a likelihood that the feature is an overlay.
- The processor determining that this assigned probability exceeds a threshold value.
- Responsive to this determination, the processor performing a specific action, selected from a list that includes determining advertising compliance, performing OCR, or determining a channel associated with the feature.
- The complaint notes that it reserves the right to assert dependent claims. (Compl. ¶27).
III. The Accused Instrumentality
Product Identification
- The complaint identifies the "Infringing Products and Methods" as the audience measurement systems and services offered by HyphaMetrics, which utilize a "coreMeter" box. (Compl. ¶29-30).
Functionality and Market Context
- The accused systems are alleged to perform audience measurement by connecting a "coreMeter" box to televisions in participating households. (Compl. ¶29). Functionally, the system is alleged to examine video feeds for "broadcaster logos" to identify content. (Compl. ¶31). The complaint alleges, based on Defendant's own patent and public statements, that this detection is a "machine-learned process" that yields a "probabilistic output" indicating the probability of a logo's presence, which is then used to determine the channel being viewed. (Compl. ¶34-35). These services are positioned to compete with Nielsen's own audience measurement products and are allegedly used as a "currency to plan, transact and measure national media campaigns." (Compl. ¶38).
IV. Analysis of Infringement Allegations
'588 Patent Infringement Allegations
| Claim Element (from Independent Claim 1) | Alleged Infringing Functionality | Complaint Citation | Patent Citation |
|---|---|---|---|
| a processor applying a feature map network to an image to create a feature map, wherein the feature map comprises a grid of vectors characterizing at least one feature in the image | The accused products allegedly use machine learning, specifically convolutional neural networks (CNNs), which function as a feature map network to process an image and create a feature map consisting of a two-dimensional array of vectors. (Compl. ¶43-44). | ¶43-44 | col. 7:50-55 |
| the processor applying a probability map network to the feature map to create a probability map assigning a probability to the at least one feature in the image, wherein the assigned probability corresponds to a likelihood that the at least one feature is an overlay | The accused products' machine-learning modules allegedly produce a "probabilistic output" indicating the probability of a network logo's presence, which the complaint equates to the claimed probability map. (Compl. ¶45, citing ¶35). | ¶45 | col. 9:25-31 |
| the processor determining that the assigned probability exceeds a threshold probability value | Upon information and belief, the complaint alleges that the accused products detect logos by determining that the "probabilistic output" for a feature exceeds a threshold value. (Compl. ¶46). | ¶46 | col. 11:7-9 |
| responsive to the processor determining that the assigned probability exceeds the threshold probability value, the processor... (iii) determining a channel associated with the at least one feature... | The accused products allegedly analyze a detected logo to "determine what channel the viewer is watching," directly corresponding to the claimed action of determining an associated channel. The complaint references a flowchart from Defendant's own patent to support this allegation. | ¶47 | col. 12:20-21 |
Identified Points of Contention
- Technical Questions: A central technical question will be whether the accused HyphaMetrics system employs the specific two-stage architecture recited in the claims—a "feature map network" followed by a distinct "probability map network." The complaint alleges this structure by citing general technical articles and the defendant's own patent, but the actual architecture of the accused product will be a point of discovery and potential dispute.
- Scope Questions: The infringement theory relies heavily on statements in HyphaMetrics' own '002 patent to describe the accused product's functionality. This raises the question of whether the '002 patent is an accurate and complete description of the accused products as sold, or if there are material differences between what the patent describes and how the commercial products actually operate.
V. Key Claim Terms for Construction
The Term: "feature map network"
- Context and Importance: The infringement analysis hinges on whether the accused system's initial processing stage meets the definition of a "feature map network." Practitioners may focus on this term because the defendant could argue its system uses a single, undifferentiated neural network, rather than the distinct two-network architecture implied by the claims.
- Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The specification provides a functional description, stating the network "receives as an input at least one image... and outputs a feature map." ('588 Patent, col. 6:39-43). This could support a construction not limited to a specific architecture.
- Evidence for a Narrower Interpretation: The patent's primary embodiment describes a very specific 12-layer convolutional neural network derived from the VGG-16 model. ('588 Patent, FIG. 8; col. 8:15-34). This could be used to argue the term is implicitly limited to this disclosed structure.
The Term: "probability map network"
- Context and Importance: This term's construction is critical for the same reason as "feature map network"; it defines the second required component of the claimed system. The dispute will likely center on whether the accused product has a separable component that performs this function.
- Intrinsic Evidence for Interpretation:
- Evidence for a Broader Interpretation: The specification functionally describes it as a network that "receives as an input the feature map" and "creates a probability map... corresponding to the likelihood that particular pixels... are part of an overlay." ('588 Patent, col. 6:58-64).
- Evidence for a Narrower Interpretation: The detailed description and preferred embodiment disclose a specific architecture using two recurrent layers (bi-directional LSTMs) that process the feature map vertically and horizontally. ('588 Patent, FIG. 9; col. 9:56-10:4). This may support an argument that the term requires a recurrent neural network structure.
VI. Other Allegations
Willful Infringement
- The complaint alleges willful infringement based on knowledge of the '588 patent acquired from the filing of the lawsuit itself. (Compl. ¶42, ¶52). No allegations of pre-suit knowledge are made.
VII. Analyst’s Conclusion: Key Questions for the Case
- A core issue will be one of architectural mapping: can Nielsen prove that the accused HyphaMetrics system embodies the specific two-stage architecture of a "feature map network" followed by a separate "probability map network," as required by the claims, or will HyphaMetrics successfully argue its system is a single, integrated process that falls outside the claim scope?
- The case will also turn on a key evidentiary question: to what extent can Nielsen rely on HyphaMetrics' own '002 patent as an admission that describes the functionality of the accused products? The court will need to determine if the '002 patent is a sufficient proxy for the accused system's actual operation or if direct evidence from the product itself reveals a material disconnect.
- Finally, the outcome may depend on claim construction: will the court define key terms like "feature map network" broadly based on their functional descriptions in the specification, or will it narrow their scope to the specific VGG-based and bi-directional LSTM embodiments that the patent discloses as the preferred implementation?