Local Image Context Descriptor Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object and scene recognition technologies on mobile devices rely on remote servers for image analysis, compromising user privacy and not providing rich contextual information, as they require continuous data transfer and scrutiny of user location or voice commands.
Innovation Solution
A system that processes images locally on portable digital devices using convolutional neural networks to generate context descriptors, allowing for real-time interpretation of user context without transferring pixel information, enabling discrete and rich keyword-based analytics and user experience enhancements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If image recognition is performed on remote servers, then processing capability is improved, but user privacy is compromised and data transfer requirements increase
Solution Approach 1:
The patent extracts the essential function of image recognition from remote servers and implements it locally on mobile devices. The system processes images on-device using machine learning models, extracting only the necessary contextual information (scene type, objects detected) without transmitting actual image data to external servers. This resolves the contradiction by maintaining processing capability while eliminating privacy concerns associated with remote image analysis.
Solution Approach 2:
The patent introduces an on-device machine learning model as an intermediary between the camera and any potential remote analysis. This local model processes images first, extracting contextual descriptors that can then be used locally or transmitted as anonymized metadata. The intermediary ensures that raw image data never leaves the device, thereby maintaining privacy while still enabling sophisticated image understanding.
2Extent of automation
If continuous image analysis is performed, then contextual awareness is improved, but energy consumption increases
Solution Approach 1:
The patent implements periodic image capture and analysis rather than continuous processing. The system captures images at specific intervals or triggered by certain events (e.g., user actions, scene changes), processes them locally using energy-efficient neural networks, and updates contextual awareness only when necessary. This approach maintains effective contextual awareness while significantly reducing energy consumption compared to continuous analysis.
Solution Approach 2:
The patent applies partial action by processing only the most relevant features of images rather than analyzing every pixel in detail. The on-device machine learning models are optimized to extract key contextual information (scene type, major objects) without performing exhaustive analysis, thereby achieving sufficient contextual awareness with reduced computational and energy requirements.
3Loss of information
If detailed scene information is collected, then analytics quality is improved, but data transmission requirements increase
Solution Approach 1:
The patent extracts only the essential contextual descriptors from images (e.g., scene type, detected objects, key attributes) and transmits only this extracted information rather than raw images or comprehensive scene data. This extraction approach maintains high analytics quality by preserving the most relevant information while minimizing data transmission requirements to a fraction of what would be needed for full image upload.
Data Source
AI summary
A method of analysing an image the image transmitted from a local image acquisition device to a local image processor; the local image processor processing the image locally in order to define at least one context descriptor relevant to a scene contained in the image.