[postgis-users] RE: PostGIS WKT Raster

[Marshall, Steve]

Pierre,

Regarding your question:
-What is the status of PGRaster? Is any development now underway?

The short answer is no.

I decided to implement our raster-in-a-database solution differently
than what I proposed in PGRaster.  When I began, it was not clear that I
would be able to stay with the PostgreSQL/PostGIS platform, so I opted
to use a more purely-relational approach that did not rely on
PostgreSQL-specific capabilities.   Thus, I have not continued on with
PGRaster.

In my new model, I stored the raster data such that each logical image
is represented by one record in an image tables, with actual data in 1:M
corresponding records in a tile table.  Each tile is formatted as a PNG
image and stored in a bytea column.  There is also an associated mapping
(called a pixel-value map) that lets one translate pixel values into
numeric values in some unit of measure, and also deal with missing data
values.  This allows the images to represent real-world (e.g. measured)
quantities.

The only integration with PostGIS is that the bounding box of each
logical image and its tiles are represented as PostGIS geometry.  There
are no built-in functions to operate on this raster data, so it probably
does not meet your needs.

In my personal opinion, my "half-way" integration of raster data into an
RDBMS is not particularly useful.  It provides a way to store and
retrieve the data, but a file-based system would be faster, and probably
more desirable.  

The biggest problem I found with storing rasters in a RDBMS is
performance.  In general, most API for raster data assume you are
working with files, and exploit this fact to get good performance with
large data sets.  If you try to represent rasters within PostGIS,
performance issues are likely to be one of your biggest problems.

I also found it hard to get application programmers interested in coding
to the relational API.  You will be asking these programmers to give up
almost all the image processing toolkits they already use, so you will
need to add quite a bit of functionality to make your solution more
attactive to them than the status quo. Howeer, integration of
RDBMS-rasters with a rich set of processing functions might make this
approach more attactive.

Since I am not continuing work on PGRaster, I probably won't be much
help to you.  However, I wish you the best of luck in this endeavor.  

Yours,
Steve Marshall