PTAB

IPR2019-01475

Robert Bosch LLC v. Monument Peak Ventures LLC

Log In
Key Events
Petition
petition

1. Case Identification

2. Patent Overview

  • Title: Method for Detecting Objects in Digital Images
  • Brief Description: The ’461 patent discloses a method and system for detecting objects, such as faces, in a digital image. The core concept involves processing an image along two parallel paths to generate two different segmentation maps: a first map generated according to a "non-object specific criterion" (e.g., general color homogeneity) and a second map generated according to an "object specific criterion" (e.g., skin color associated with a face). Both maps are then used to detect the target object.

3. Grounds for Unpatentability

Ground 1: Anticipation of Claims 1, 2, and 15-17 by Lestideau

  • Prior Art Relied Upon: Lestideau (Application # 2003/0053685)
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Lestideau, which discloses a method for locating human faces, teaches every limitation of the challenged claims. Independent claim 1 requires generating a first segmentation map using a non-object specific criterion and a second map using an object-specific criterion, then detecting objects using both. Petitioner asserted that Lestideau’s step of creating a "binary texture map" to identify high-textured areas (where skin is not typically found) corresponds to the claimed first segmentation map based on a non-object specific criterion (texture). Petitioner further asserted that Lestideau’s "skin colour analysis" to detect segments with a high probability of being skin corresponds to the second segmentation map based on an object-specific criterion (skin, for the object "face"). Finally, Petitioner contended that Lestideau detects faces by using both maps, specifically by using boundary boxes derived from the texture map to analyze the skin segments, thus satisfying the final limitation of the independent claims. For claim 2, Petitioner argued that Lestideau’s use of boundary boxes from the texture map to constrain the analysis of skin segments constitutes "merging" the two maps prior to detection. The arguments for system claims 16 and 17 mirrored those for method claims 1 and 2.

Ground 2: Anticipation of Claims 1, 9, 15, and 16 by Kinjo

  • Prior Art Relied Upon: Kinjo (Patent 5,978,100)
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Kinjo, which discloses detecting facial regions, also anticipates the challenged claims. Petitioner mapped Kinjo's "Background-Portion Eliminating Method"—which identifies and removes regions of a particular color like blue sky or green turf—to the claimed "first segmentation map" based on a "non-object specific criterion" (a general background color not specific to the target object). Petitioner mapped Kinjo's "Face-Candidate Extracting Method"—which searches for "patterns of shape peculiar to various parts of a human figure"—to the "second segmentation map" based on an "object specific criterion" (human-specific shapes). Kinjo's system determines a final facial region "on the basis of the results of processing by the extracting units" (plural), which Petitioner argued means using both the non-object specific and object-specific maps for detection. For dependent claim 9, which adds segmenting into regions of homogeneous color and assigning a probability score, Petitioner asserted that Kinjo’s methods of segmenting by flesh color and assigning a "weighting coefficient" to determine the highest probability region met these limitations.

4. Key Claim Construction Positions

  • "non-object specific criterion" / "object specific criterion": Petitioner proposed construing "non-object specific criterion" as "a criterion that does not match image pixels to a specific object identified within the digital image." Conversely, "object specific criterion" was proposed as "a criterion that matches image pixels to a specific object identified within the digital image." Petitioner argued that the patent's own description distinguishes between a general property like skin color (which it uses in both processing paths) and a true object-specific criterion that involves additional properties like shape, as described in the "Body Plans" prior art cited by the patent itself.
  • "stage": Petitioner argued that in the context of the ’461 patent, "stage" refers to "a portion of a software program or electronic circuit" that performs a claimed function. This interpretation avoids it being a "nonce" word. Alternatively, should the Board find the term invokes pre-AIA 35 U.S.C. §112, ¶ 6 (means-plus-function), Petitioner identified the corresponding structures in the specification for each "stage" element, such as "color segmentation stage 12" for the first stage and "object specific path 16" for the second stage.
  • "a detector for detecting objects...": Petitioner contended this term is a means-plus-function element under pre-AIA §112, ¶ 6. The claimed function is "detecting objects in the digital image using both first and second segmentation maps." The corresponding structure disclosed in the specification was identified as a processor programmed to perform one of two algorithms: (a) merging the maps and then detecting objects in the merged map, or (b) detecting objects in each map separately and then merging the results.

5. Relief Requested

  • Petitioner requested institution of an inter partes review and cancellation of claims 1, 2, 9, and 15-17 of the ’461 patent as unpatentable under 35 U.S.C. § 102.