Return-Path: pg_adm@postgres.berkeley.edu
Received: by postgres.Berkeley.EDU (5.61/1.29)
	id AA25139; Wed, 28 Jul 93 07:30:32 -0700
Date: Wed, 28 Jul 93 07:30:32 -0700
Message-Id: <9307281430.AA25139@postgres.Berkeley.EDU>
From: payton@dipl.rdd.lmsc.lockheed.com (Paul Payton x42114)
Subject: Multi-media Databases
To: postgres@postgres.berkeley.edu
Sender: pg_adm@postgres.berkeley.edu
Cc: payton@dipl.rdd.lmsc.lockheed.com

	By way of introduction, my name is Paul M. Payton and I'm working in a
research lab with a small group of people handling images. We have a desire
to acquire a commercial or quality-grade 'freeware' package which will support
the storage of multimedia data types. Initially, we are focused on storing
images, imagery-related data (photogrammetric information, polynomial warp
coefficients), and imagery-derived data (chained edges from Canny operations
and polygonal delineations from segmentation algorithms) in a common database
with sophisticated query capabilities (spatial, temporal, and GIS-like on top of
the usual table joins seen in RDBMSs).

	We have a copy of POSTGRES here in our facility and feel it has many
of the properties we like. However, as we all know, it has some limitations and
some bugs still being exterminated. And I wouldn't feel justified in strongly
suggesting it as the basis for a deliverable product.

	I am interested in learning of any commercial-grade products which
support the traditional relational operations of RDBMS along with graphical,
text, audio, and imagery data types. Additionally, I would like a product which
has GIS functionality built-in to the system, so that containment, overlap,
and adjacency conditions could be imposed on the 'where' clause of a standard
sort of SQL query. We could, for example, pull out all the linear cultural
features from a DMA DFAD set of tables which overlap a polygonal set of image
coverage coordinates in WGS84 standard latitude/longitude frame.

	ERDAS, GRASS, etc. are great for geo-referenced data types (mainly
images and rasterized/vectorized data) but do not support the RDBMS part of the
problem we're trying to tackle.

	ITASCA has a Lucid LISP-based product with C and C++ user bindings
for embedded calls to the database (actually a LISP image). With an executable
of size > 15MB which immediately asks the OS for 40MB more space, well...

	Servio-Logic has a product which, at last I heard, interfaces to a
Smalltalk "database" (more a network). In product demonstrations I've seen, it
seemed to act flaky and run slowly. (But that was a while ago.)

	UNISQL is a product I've heard mumblings about and, from experts around
here, might fulfill my shopping list of needs.

	POSTGRES seems to be the closest to the mark I can find. I have heard
mention of MIRO, a commercial product of the same lineage and wonder what
experiences users have had with it. It's flexible, which I like. And we could
add our own custom GIS functions AND rules/triggers/agents. But, would you trust
it to run in your customer's facility for weeks on end?

	Symbolics has some object-oriented software (Statice, I think?) which
runs on the Common Lisp platform. Nice development environment -- but not the
right one for our applications.

	My primary interest is a C/C++ native product (as little binding to
foreign stuff like LISP as possible) which runs on UNIX/ULTRIX.

	I'm sure there's no dream product out there...but I would like to pick
the best foundation to build something clean and modular.

	Any pointers or suggestions or caveats or pithy words?

-----------------------------------
Paul M. Payton - Computer Scientist
Lockheed Missiles and Space Company
Research and Development Division (Palo Alto)
E-Mail: payton@dipl.rdd.lmsc.lockheed.com
V-Mail: (415) 354-5753
(Fewer "Missiles" and more "Space")