This is an experiment - in order to evaluate the feasibility of using SQLite as the persistence-layer (or file-format) for multi-dimensional images (as encountered e.g. in microscopy, cf. OME-TIFF or BioFormats).
The images we are dealing with are organized as large mosaics with a pyramid.
For a viewer capable of fast zooming and panning, it is important to quickly retrieve the tiles intersecting with the specific viewport. For this task the R*Tree module of SQLite seems well suited and is employed.
The bitmap data itself is being stored inside the SQLite-database (as a BLOB). This is a very controversial approach, and there is a lot of discussion what the best architecture for dealing with large blobs in databases is (e.g. here, here). So, getting some real-world experience here is one goal of this experiment.
The same techniques (ie. image pyramids and tiling) are applied to 3D-volumetric data.
This library focuses solely on the storage aspect; image processing aspects are out-of-scope and they are expected to be implemented on another layer.
Currently, two data models are implemented and available via the interfaces IDbRead/IDbRead3D (and IDbWrite/IDbWrite3D)
- for multi-dimensional, tiled 2D-images with pyramid
- and for multi-dimensional, tiled 3D-images (refered to as bricks hereafter) with pyramid
In both cases, the tiles/bricks are paraxial.
A tile/brick is described by
| field | description |
|---|---|
| logical position | coordinates and width/height of the tile |
| tile coordinate | an arbitrary sized attribute |
The tile coordinate are attributes with a specific number of dimensions, allowing to group the tiles.
All tiles are positioned in X-Y-direction in a common cartesian coordinate system. In case of bricks the logical position includes a Z-coordinate and a depth.
Why should image data be stored in a database? What is wrong with traditional file-formats like TIFF?
Some of the advantages expected from using RDBMS (and SQLite in particular) are:
- consistent and unambigous data structure
- The database enables us to have the data indexed so that queries are fast. In particular, those indexes are managed by the database and they are persistent.
- Query operations can be formulated in SQL.
- Transparency and self-documentation - the data is easily accessible and the storage format is future proof (see here).
- Relying on tested-and-true technology.
Why is the code in C++? Wouldn't C#, Python, Javascript, Go, D,... be preferable?
Short answer - no. Main reasons being that interfacing with SQLite is best done in C/C++; C++ is mature, time-tested and offers excellent performance and it is portable to literally any platform. It is expected that bindings for languages of your choice can be built on top of the C++-layer.
TODO
TODO
TODO
TODO
We use SemVer for versioning. For the versions available, see the tags on this repository.
- Jürgen Bohl - Initial work - [email protected]
This project is licensed under the GPL3 License - see the LICENSE file for details