Image Encapsulation Metadata for Tile-Based Region Extraction

The patent segments the image data into tiles, where each tile can be independently accessed and processed. The media container is divided into multiple boxes (e.g., 'mdat' for media data, 'moov' for metadata, 'stbl' for sample tables) that organize different aspects of the image data. This segmentation enables selective access to specific tiles without processing the entire image, improving ease of operation while maintaining manageable complexity through standardized box structures.

The patent implements a nested box structure within the media container, where metadata boxes contain references to sample tables, which in turn reference individual tile data. This nested organization allows the system to manage complexity by hierarchical structuring - the outer boxes contain high-level descriptions while inner boxes contain detailed tile information, enabling efficient access without requiring parsing of the entire file structure.

The patent performs preliminary organization of image data into tiles during the encoding process, with tile boundaries and descriptions being established in advance. The sample table and metadata boxes are pre-configured with references to specific tile locations and descriptions. This preliminary action eliminates the need for post-processing analysis to identify regions of interest, as the tile structure is already in place, thereby improving extraction efficiency without requiring excessive metadata to be added later.

The patent designs the tile description structure to serve multiple functions: the same tile metadata structure supports various applications including region of interest extraction, computational photography operations, and video summarization. The sample table and box references are designed to be universally applicable across different use cases, reducing the need for application-specific metadata and minimizing overall metadata quantity while maintaining high productivity across diverse tasks.

The patent segments video data into temporal slices and spatial tiles, creating a two-dimensional division where each tile corresponds to a specific time frame and spatial region. This segmentation allows the system to maintain high quality for selected sub-parts by processing only relevant tiles rather than the entire video stream. The standardized box structure for organizing these segmented units simplifies the processing complexity by providing clear demarcation and reference points for each tile.

The patent implements a dynamic tile selection mechanism where the system can adaptively choose which tiles to process and transmit based on user interaction and content importance. The sample table structure enables dynamic referencing of specific tiles without requiring reprocessing of the entire video. This dynamic approach maintains manufacturing precision for quality-critical regions while reducing overall processing complexity by focusing computational resources only on selected tiles rather than the complete video sequence.