Neural Network Filter for Photosensitive Media Safety
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Photosensitivity, an adverse reaction to light stimuli, affects millions worldwide, particularly children, and existing TV broadcast standards only address a small subset of stimuli, leaving many at risk for seizures and discomfort due to unfiltered strobing and rapidly-changing patterns.
Innovation Solution
A machine learning-based system that detects and modifies adverse visual stimuli in media, such as videos, by training neural networks to remove or attenuate flashes and patterns, using datasets with added artifacts to learn inverse transformations, applicable in screen filters, augmented reality glasses, and content providers to ensure safety for photosensitive individuals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If TV broadcast standards limit certain types of strobing, then protection against red-blue flashes is improved, but other types of adverse stimuli (white-black flickers, rapidly-changing patterns) remain unaddressed
Solution Approach 1:
The patent implements a universal filtering system that handles multiple types of adverse stimuli (red-blue flashes, white-black flickers, rapidly-changing patterns) through a single comprehensive solution. The machine learning model is trained to detect and filter various stimulus types with different characteristics, making the system adaptable to diverse photosensitivity triggers rather than requiring separate filters for each stimulus type.
Solution Approach 2:
The system dynamically adjusts filtering parameters based on the detected stimulus characteristics. Different stimulus types (flashes, flickers, patterns) have different ranges and frequencies, and the filter adapts its attenuation parameters accordingly. The machine learning model learns to identify stimulus parameters such as frequency, duration, and spatial distribution to apply appropriate filtering strength for each type.
2Reliability
If existing filters attenuate flashes, then protection against flash stimuli is improved, but filters cannot address rapidly-changing patterns that cause similar adverse reactions
Solution Approach 1:
The filtering system is designed to handle both flash stimuli and rapidly-changing patterns through a unified machine learning approach. The model detects temporal and spatial characteristics of different stimulus types and applies appropriate filtering, making the system versatile enough to protect against multiple categories of adverse stimuli rather than being limited to flash attenuation only.
3Reliability
If regulators institute TV broadcast standards, then protection against regulated stimuli is improved, but the standards address only a small subset of stimuli that cause seizures
Solution Approach 1:
The system extends protection beyond regulated stimuli by dynamically detecting and filtering unregulated adverse stimuli based on their physical parameters. The machine learning model identifies characteristics such as frequency, contrast, and temporal patterns that indicate potential seizure triggers, allowing the system to adapt to both regulated and unregulated stimulus types with a single comprehensive approach.
Data Source
AI summary
In an embodiment, a method, and corresponding system and non-transitory computer readable medium storing instructions configured to cause a processor to execute steps are configured to introduce an auxiliary transformation to a digital media, resulting in a transformed digital media by generating the auxiliary transformation with a transform function. The method is further configured to evaluate the transformed digital media to generate a metric estimating a human response to the transformed digital media altered by the introduced auxiliary transformation. The method is further configured to train a neural network to remove the auxiliary transformation from any digital media by learning a desired transformation function from the transformed digital media and the metric associated with the transformed digital media.