Video Filtering Device Using Previous Frame Metadata Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current AI technologies lack an effective method to filter out harmful or undesired content from video playback, such as violent, lascivious, or horror scenes, which can be distressing for users.

Innovation Solution

A device and method that predict the harmfulness of upcoming video frames based on metadata generated from previous frames, using image, speech, and text information, and filter out harmful content by comparing predicted harmfulness levels with user-set filtering levels, adjusting visibility accordingly.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If deep learning is used to filter harmful images in video, then the recognition accuracy is improved, but the processing time and computational complexity increase

Engineering Contradiction:
Improverecognition accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary action by analyzing previous frames before the current frame to predict harmfulness of upcoming frames. This allows the filter to be proactive rather than reactive, reducing the need for complex real-time analysis of each frame while maintaining high accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The video filtering process is segmented into distinct stages: analyzing previous frames, generating metadata, predicting harmfulness of next frames, and making filtering decisions. This segmentation allows complex deep learning operations to be distributed across multiple simpler processing steps, reducing overall computational burden.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If frame-by-frame analysis is performed to accurately detect harmful content, then the filtering precision is improved, but the processing speed decreases

Engineering Contradiction:
Improvefiltering precisionVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

By performing analysis on previous frames in advance, the system prepares metadata that can be quickly referenced when making filtering decisions for current and next frames, avoiding the need for slow frame-by-frame analysis at the moment of display.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system creates metadata copies from previous frames that capture essential information about harmful content patterns. These metadata copies can be quickly compared and processed without requiring the full computational resources needed for complete frame analysis.

Inventive Principle:
Principle #26Copying

3Adaptability or versatility

If multiple categories of harmfulness are predicted for each frame, then the comprehensive detection capability is improved, but the computational load increases

Engineering Contradiction:
Improvedetection capabilityVSAvoidcomputational load
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The harmfulness detection is segmented into multiple independent category assessments (e.g., violence, horror, lasciviousness). Each category can be evaluated separately using specialized metadata from previous frames, allowing the system to maintain high detection capability while managing computational load through modular processing.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4280612A1Method and apparatus for filtering video
Publication Date: 2023.11.22 SAMSUNG ELECTRONICS CO LTD
  • EP4280612A1 patent drawingFigure 1
  • EP4280612A1 patent drawingFigure 2
  • EP4280612A1 patent drawingFigure 3

AI summary

An artificial intelligence (AI) system for simulating functions such as recognition, determination, and so forth of a human brain by using a mechanical learning algorithm such as deep learning, or the like, and an application thereof are provided. A method of filtering video by a device is provided. The method includes selecting at least one previous frame preceding a current frame being played from among a plurality of frames included in the video, generating metadata regarding the selected at least one previous frame, predicting harmfulness of at least one next frame to be displayed on the device after playback of the current frame, based on the generated metadata, and filtering the next frame based on the predicted harmfulness.