PTAB

IPR2025-01522

Samsung Electronics America, Inc. v. SnapAid, Ltd.

Log In

1. Case Identification

2. Patent Overview

  • Title: Real Time Assessment of Picture Quality
  • Brief Description: The ’682 patent describes a computerized method for assessing the quality of a captured image. The system uses various sensors (e.g., motion, location) and image analysis to determine multiple quality indicators (QIs), which are then weighted and combined to calculate a total quality score and provide real-time feedback.

3. Grounds for Unpatentability

Ground 1A: Obviousness over Core Image Processing References - Claims 1, 2, 5-8, 11, and 14-16 are obvious over Anon in view of Takeuchi and Jasinski.

  • Prior Art Relied Upon: Anon (Patent 8,508,622), Takeuchi (Application # 2010/0149361), and Jasinski (Application # 2012/0201427).
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Anon taught the foundational method of claim 1: an image capturing device that determines image-based characteristics (e.g., focus, brightness), computes a "composition measure" using a weighted combination of parameters, and provides real-time feedback. However, to the extent Anon did not explicitly teach dynamically estimating weights based on image context, Jasinski supplied this limitation by describing a system that computes weights for image regions dynamically for each new image, allowing adaptation to changing scene conditions. Furthermore, Takeuchi was argued to disclose analyzing a captured image to detect objects (e.g., faces), assess their motion using motion vectors, and evaluate aesthetic quality, thereby teaching the limitations related to obtaining third (QI3), fourth (QI4), and fifth (QI5) quality values.
    • Motivation to Combine: A POSITA would combine Anon with Takeuchi to enhance Anon’s general quality assessment with more specific, context-aware inputs like subject detection and motion, thereby improving the relevance of the feedback. A POSITA would further incorporate Jasinski’s dynamic weighting as a known technique to improve the accuracy of such systems by allowing the quality assessment to adapt to current scene conditions, which is a natural extension of Anon's weighted-parameter approach.
    • Expectation of Success: Petitioner asserted a POSITA would have had a reasonable expectation of success because combining these modular, complementary techniques was a predictable way to improve image quality assessment. The integration involved applying known methods (object detection, dynamic weighting) to a known system (real-time quality feedback) to achieve a more robust and accurate result.

Ground 1B: Obviousness over Core References and Aisaka - Claims 17, 19, and 20 are obvious over Anon in view of Takeuchi, Jasinski, and Aisaka.

  • Prior Art Relied Upon: Anon (Patent 8,508,622), Takeuchi (Application # 2010/0149361), Jasinski (Application # 2012/0201427), and Aisaka (Application # 2010/0246939A1).

  • Core Argument for this Ground:

    • Prior Art Mapping: This ground challenged independent claim 17, which requires analyzing an image via "deep learning algorithms." Petitioner relied on the combination of Anon, Takeuchi, and Jasinski as in Ground 1A to teach the majority of the claim limitations, including obtaining quality values for device motion (QI1) and camera exposure/focus (QI2) and calculating a total quality value. To supply the deep learning element, Petitioner introduced Aisaka, which taught an image processing apparatus that uses "neural-network-based learning" to detect subjects, generate a "subject map," and calculate a score reflecting the subject's quality (e.g., sharpness). This score was argued to be the claimed third value (QI3) obtained via deep learning.
    • Motivation to Combine: A POSITA would be motivated to integrate Aisaka’s deep learning methods into the system of Anon and Takeuchi to improve the accuracy and robustness of subject detection and evaluation. This would be a logical improvement over the pattern matching and rule-based methods of the primary references, directly aligning with the shared goal of providing more reliable and automated image quality assessment.
    • Expectation of Success: Petitioner contended that success would be expected because deep learning-based object recognition was an established and effective technique for image analysis at the time. The frameworks of Anon and Takeuchi were described as modular and readily adaptable to incorporate more advanced algorithms like those taught by Aisaka.

  • Additional Grounds: Petitioner asserted additional obviousness challenges against various dependent claims by augmenting the core combination of Anon, Takeuchi, and Jasinski with other references. These included combinations with:

    • Alhadef (Patent 8,009,198) for compensating for sensor drift and error (Claim 3).
    • Ramesh (Application # 2009/0296989) for probabilistic error estimation in object detection (Claims 4, 9).
    • Steinberg (Patent 8,682,097) for calculating statistical measures like mean and standard deviation over time (Claim 10).
    • Liu (an IEEE 2012 publication) for fusing scores from multiple aesthetic algorithms (Claim 12).
    • Cheatle (Application # 2002/0110286) for using a pre-stored list of faces for recognition (Claim 13).
    • Yang (Application # 2011/0222724) for clarifying the use of deep learning (Claims 14, 15, 19, 20).
    • Ishiwata (Application # 2008/0013851) for combining accelerometer, gyro, and lens data for camera motion (Claim 18).

4. Relief Requested

  • Petitioner requests institution of an inter partes review and cancellation of claims 1-20 of the ’682 patent as unpatentable under 35 U.S.C. §103.