Efficient multi-dimensional storage of very large dense, sparse and semi-sparse image datasets with machine learning
Inventors
Fedorov, Dmitry • Lang, Christian A. • Kvilekval, Kristian • Goldberg, Ilya
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12536608-B2
Publication Date
2026-01-27
Expiration Date
Abstract
Methods, apparatus, and systems for high-resolution image storage and analysis are disclosed. In one disclosed embodiment, a method includes analyzing a high-resolution image; partitioning the high-resolution image into logical blocks; detecting objects in the high-resolution image within the logical blocks, assigning objects that differ or with differing measures to differing layers, and assigning objects that are similar or with similar measures to the same layer; tabulating the detected objects into a table, wherein each row indicates a spatial position of each detected object and at least one measure of each detected object; storing the spatial position and the at least one measure of each detected object into the table; spatially rendering selected objects over the high-resolution image; and displaying the selected objects at a selected resolution within a portion of the high-resolution image on a display device.
Core Innovation
The invention relates to storing and processing a large-scale digital image by reading the digital image within an image space at a one hundred percent scale, where the one hundred percent scale represents the highest resolution of the digital image. The image space is defined using spatial parameters, measure parameters, and resolution parameters, and is partitioned into a plurality of logical blocks that are arbitrarily positioned and arbitrarily sized, with at least some blocks overlapping on one or more areas of the image space.
Imaged objects and associated measures are detected in each logical block by using the measure parameters. The detected imaged objects associated with the logical blocks are stored at the one hundred percent scale and one or more lower resolutions, including tabular storage of the associated measures, object types, and object identifiers into a table, and storage of object spatial data into a semi-sparse matrix associated with the object identifiers.
The approach supports multiple resolutions and scalable processing by combining overlapping logical blocks with resolution-dependent storage representations. The document also describes optimizing storage by fusing overlapping parts of imaged objects identified in overlapping blocks, generating further lower resolutions by densifying and re-gridding the digital image for fast visualization, and machine learning inference to iteratively add new measures while maintaining the linked object identifier-based storage structure.
Claims Coverage
The partial content includes two independent claims, one method claim and one system claim. Both claims cover four inventive features in a shared storage and processing framework using overlapping logical blocks, detecting imaged objects with predefined metrics, and storing linked tabular and semi-sparse spatial representations across one hundred percent and lower resolutions.
Overlapping logical block partitioning at one hundred percent scale
Partitioning the image space into a plurality of logical blocks that are arbitrarily positioned and arbitrarily sized, wherein at least some of the logical blocks overlap on one or more areas of the image space.
Measure-parameter object detection within logical blocks
Detecting the imaged objects and associated measures in each of the plurality of logical blocks by using the measure parameters.
Tabular measure storage linked by object identifiers plus semi-sparse spatial matrix
Storing the imaged objects associated with the logical blocks at the one hundred percent scale and one or more lower resolutions, including tabular storage of the associated measures, object types, and object identifiers into a table, and storage of object spatial data into a semi-sparse matrix associated with the object identifiers.
Image space definition using spatial, measure, and resolution parameters including a plurality of resolutions
Defining the image space of the digital image by using spatial parameters, measure parameters, and resolution parameters, wherein the spatial parameters define the extent of the digital image and the resolution parameters define a plurality of resolutions including the highest resolution.
Across the independent method and system claims, the coverage centers on defining an image space with spatial, measure, and resolution parameters, partitioning it into overlapping arbitrarily sized logical blocks, detecting imaged objects and measures within each block, and storing linked representations at one hundred percent and lower resolutions using both tabular measure storage and semi-sparse spatial data associated with the object identifiers.
Stated Advantages
Major reductions in file size.
Faster machine-learning-derived measure generation.
Faster visualization of image space at selected resolution levels.
Documented Applications
Web/scalable image analysis platform for very large high-resolution images, including microscopy/spectroscopy imagery and aerial and satellite imagery.
Interested in licensing this patent?