PTAB

IPR2018-00688

Nikon Corp v. Carl Zeiss AG

Log In
Key Events
Petition
petition

1. Case Identification

2. Patent Overview

  • Title: Image Pattern Detection and Recognition for Human Faces
  • Brief Description: The ’163 patent describes methods and systems for detecting patterns, particularly human faces, in digital images. The core invention is a two-step "pre-filtering" process designed to efficiently identify candidate regions in an image that likely contain the target pattern before passing them to a more computationally intensive final detector.

3. Grounds for Unpatentability

Ground 1: Obviousness over Yow - Claims 1-3, 6, 9-10, 14-16, and 19 are obvious over Yow

  • Prior Art Relied Upon: Yow (a 1997 article by Kin Choong Yow and Roberto Cipolla titled "Feature-based human face detection").
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Yow taught every element of the challenged claims. Yow described a two-stage "preattentive" process for face detection that directly corresponded to the claimed pre-filtering method. The first stage involved correlating an input image with a filter (a second derivative Gaussian filter) and identifying points of "local maxima" as initial interest points, which Petitioner mapped to the claimed "correlating" step to find candidate regions. The second stage involved examining the regions around these interest points using edge detection to analyze grayscale characteristics (like gray-level variance and luminance gradients), which mapped to the claimed "screening" step based on a grayscale characteristic.
    • Key Aspects: This ground asserted that a single prior art reference disclosed the entire inventive concept, rendering the claims obvious to a person of ordinary skill in the art (POSITA) without any need for modification or combination.

Ground 2: Obviousness over Yow in view of Crowley - Claims 4, 7, and 11 are obvious over Yow in view of Crowley

  • Prior Art Relied Upon: Yow (the 1997 article) and Crowley (a 1995 article by James L. Crowley on visual processes).
  • Core Argument for this Ground:
    • Prior Art Mapping: This ground built upon the teachings of Yow. Petitioner contended that while Yow taught the core two-step pre-filtering method, it did not explicitly disclose two features recited in certain dependent claims: (1) calculating the reference data (the filter) from a "plurality of training images" (claim 4), and (2) comparing correlation values to a "threshold correlation value" to select candidate regions (claims 7 and 11). Crowley was introduced to supply these missing elements, as it explicitly taught creating an "average face image" from a number of training images to form a correlation kernel and then selecting a face candidate only when the correlation value exceeded a specified threshold.
    • Motivation to Combine: Petitioner argued a POSITA would combine Yow and Crowley because both were in the same field of face detection. A POSITA would have recognized the benefit of using Crowley's well-known technique of generating a filter from multiple training images to improve the robustness and accuracy of the initial correlation step in Yow's system. Similarly, applying a threshold to the correlation values, as taught by Crowley, was a conventional way to reduce false positives and improve the efficiency of the overall detection process.
    • Expectation of Success: The combination was presented as a predictable substitution of known techniques. Replacing Yow’s filter with one derived via Crowley's method would predictably lead to a more accurate correlation image, and applying a threshold would predictably improve filtering efficiency.

Ground 7: Obviousness over Kosugi, Fang, and Rowley - Claims 15 and 19 are obvious over Kosugi in view of Fang and further in view of Rowley

  • Prior Art Relied Upon: Kosugi (Japanese Patent Publication Hei 5-174149), Fang (WO 96/38808), and Rowley (an applicant-admitted prior art IEEE publication from 1998).
  • Core Argument for this Ground:
    • Prior Art Mapping: This ground targeted the system claims (15, 19). Petitioner argued Kosugi disclosed the overall two-stage pre-filtering system architecture, including a "Rough Search Unit" (first filtering means), a "Detailed Search Unit" (second filtering means), and a final "Recognition Unit" (image pattern detector). However, Kosugi’s second filtering step was based on grayscale intensity differences, not contrast. Fang was introduced to teach screening based on comparing contrast values to a threshold, satisfying the limitations of the claimed "second filtering means." Finally, while Kosugi’s "Recognition Unit" was generic, Rowley was introduced as well-known, applicant-admitted prior art that taught a neural network-based face detector, supplying the structure for the claimed "image pattern detector."
    • Motivation to Combine: A POSITA would combine these references to create a more effective and accurate face detection system. They would start with Kosugi's efficient two-stage architecture and be motivated to improve its components. It would have been logical to replace Kosugi’s simple grayscale intensity screening with Fang’s more sophisticated contrast-based screening and to implement Kosugi’s generic final detector using the well-known and powerful Rowley neural network.
    • Expectation of Success: The proposed combination involved replacing individual functional modules of Kosugi's system with more advanced, but well-understood and compatible, modules from Fang and Rowley. This would have led to the predictable result of a more accurate and robust face detection system.
  • Additional Grounds: Petitioner asserted additional obviousness challenges based on the Yow Thesis (a 1998 Ph.D. thesis by the same author as the Yow article), both alone and in combination with Crowley, as well as combinations of Kosugi and Fang. These grounds relied on similar teachings and combination rationales as those detailed above.

4. Key Claim Construction Positions

  • "Correlating" / "Correlation": Petitioner proposed these terms be construed to mean "an image correlation operation between the input image and a kernel that results in a correlation image." This construction was argued to be critical, as it requires the creation of an intermediate two-dimensional image, a feature Petitioner used to distinguish the claimed invention from prior art that might only calculate a probability value without generating such an image.
  • Means-Plus-Function Terms (Claim 15): Petitioner argued that three terms in system claim 15 were means-plus-function terms under 35 U.S.C. §112, ¶ 6.
    • "first filtering means": The function is identifying and correlating image regions. The disclosed structure is a linear matched filter.
    • "second filtering means": The function is screening candidate regions by examining a grayscale characteristic. The disclosed structure is a non-linear filter programmed to analyze contrast values.
    • "image pattern detector": The function is analyzing a candidate region for verification. The disclosed structure is the neural-network face detection system described in the specification (based on the Rowley publication).

5. Relief Requested

  • Petitioner requests institution of an inter partes review and cancellation of claims 1-4, 6-7, 9-11, 14-16, and 19 of Patent 6,463,163 as unpatentable.