Scene Annotation Neural Networks for Real-Time Media Accessibility

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing audio-visual media, such as videogames and movies, lack accessibility features for individuals with disabilities, particularly those with vision impairments or color blindness, and accommodations are often time-consuming and labor-intensive to implement.

Innovation Solution

An On-Demand Accessibility System that includes modules for action description, scene annotation, color accommodation, graphical style modification, and acoustic effect annotation, utilizing neural networks to enhance accessibility by adding captions, text-to-speech, and adjusting visual styles in real-time, without altering the original media.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional image processing techniques are used for scene annotation, then implementation complexity is reduced, but annotation precision deteriorates due to inability to handle large-scale data efficiently

Engineering Contradiction:
Improveannotation precisionVSAvoidimplementation complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system segments the scene annotation task into multiple hierarchical levels: coarse-grained region proposal generation followed by fine-grained category classification. This segmentation allows the system to handle large-scale data efficiently while maintaining high annotation precision through specialized processing at each level.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent replaces traditional mechanical image processing techniques with deep learning-based neural networks. The neural network automatically learns feature representations and performs scene annotation, substituting manual feature engineering and traditional algorithms with data-driven automated processing that achieves superior precision.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If deep learning methods are applied to scene annotation, then annotation precision is improved, but system complexity and computational resource requirements increase

Engineering Contradiction:
Improveannotation precisionVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The deep learning system is segmented into modular components including region proposal networks and classification networks, each with specific functions. This modular architecture reduces system complexity by allowing independent optimization and training of each component while maintaining high overall annotation precision.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The trained neural network model serves multiple functions: it can annotate different types of scenes, handle various object categories, and adapt to different data formats. This universality reduces the need for multiple specialized systems, thereby reducing overall system complexity while maintaining high precision across diverse applications.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Measurement precision

If manual scene annotation is performed, then annotation precision can be maintained, but productivity deteriorates due to time-consuming processes

Engineering Contradiction:
Improveannotation precisionVSAvoidannotation productivity
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The system performs self-service by automatically generating scene annotations without human intervention. The neural network processes images autonomously, learning from training data and producing annotations at high speed, thereby dramatically improving productivity while maintaining precision comparable to or exceeding manual annotation quality.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

Manual annotation processes are replaced with automated deep learning-based annotation systems. This substitution eliminates the time-consuming nature of manual work while preserving high precision through the neural network's ability to learn accurate feature representations and category assignments.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

4Device complexity

If traditional algorithms are used for data processing, then device complexity is reduced, but measurement precision deteriorates in handling large-scale data

Engineering Contradiction:
Improvealgorithm complexityVSAvoiddata processing precision
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

Traditional algorithms are replaced with deep learning-based neural networks that automatically learn complex patterns and features from data. This substitution enables high-precision processing of large-scale data by leveraging the neural network's capacity to capture intricate relationships that traditional algorithms cannot detect, despite the increased computational complexity.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP3874737B1Scene annotation using machine learning
Publication Date: 2026.05.13 SONY INTERACTIVE ENTERTAINMENT LLC
  • EP3874737B1 patent drawingFigure 1
  • EP3874737B1 patent drawingFigure 2A~2C
  • EP3874737B1 patent drawingFigure 2D

AI summary

A system enhances existing audio-visual content with audio describing the setting of the visual content. A scene annotation module classifies scene elements from an image frame received from a host system and generates a caption describing the scene elements. A text to speech synthesis module may then convert the caption to synthesized speech data describing the scene elements within the image frame.