AI Metadata Storage in ISO Base Media Files
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current media file formats lack a standardized mechanism for storing and signaling artificial intelligence (AI) metadata, limiting the interoperability and usage of AI-based media manipulation services and algorithms.
Innovation Solution
The proposed solution involves defining new boxes and relationships within the ISO Base Media File Format (ISOBMFF) to store and signal AI metadata, including a new property box for descriptive metadata, a box for AI algorithm output, and a media sample definition for timed metadata tracks, enabling efficient and standard storage and signaling of AI-related data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If visual data is processed to detect content elements using AI algorithms, then semantic information about media content is obtained, but standardized storage and signaling of AI metadata is lacking
Solution Approach 1:
The patent applies universality by creating a standardized metadata container structure that can store multiple types of AI processing information (algorithm identifiers, content element detections, processing parameters) in a unified format. This universal container enables different AI systems and media players to interchange and process AI metadata consistently across various platforms and applications.
Solution Approach 2:
The patent segments AI metadata into distinct structured components including algorithm identification fields, content element detection results, and processing parameter storage. This segmentation allows for organized, modular storage of different types of semantic information, making it easier to retrieve and process specific metadata elements without handling the entire metadata set.
2Adaptability or versatility
If AI metadata is stored without standardization, then flexibility in storing different algorithm outputs is achieved, but interoperability between systems is limited
Solution Approach 1:
The patent utilizes parameter changes by defining a structured metadata container with specific fields and data types that can accommodate various AI algorithm outputs. The container includes parameters for algorithm identifiers, detection confidence levels, content element categories, and processing timestamps, allowing flexible storage of different algorithm results while maintaining consistent structure for reliable system interoperability.
3Loss of information
If multiple processing algorithms are applied to visual data, then comprehensive content analysis is achieved, but complexity of metadata management increases
Solution Approach 1:
The patent segments metadata from multiple processing algorithms into distinct, organized fields within the container structure. Each algorithm's output is captured in separate metadata entries with unique identifiers, allowing comprehensive content analysis results to be stored systematically. This segmentation reduces management complexity by providing clear organization and retrieval paths for metadata from different algorithm sources.
Data Source
AI summary
The embodiments relate to a method comprising receiving visual data in a file format compatible with ISO base media file format; processing the visual data to detect one or more content elements; storing the detected one or more content elements and information on the used process as a metadata; and including the metadata to the media file in association with the visual data. The embodiments also relate to a technical equipment for implementing the method.