Image LSID Resolution Prototypes

1
Image LSID Resolution Prototypes

• Hui Dong, Bob MorrisUMass Boston

2
Application

• Toy application at http//tamarin.cs.umb.edu8081/
  jsp-examples/MyJsp.jsp

3
Image Lsid Resolution Servers

• Projects
• UMB from local image store of 13,000 field
  photographs. (Morris, Haber, Dong)
• Morphbank (U. Florida) from project to document
  morphological characters (Rohnquist, Riccardi)
• U.T. Austin X-Ray CT facility to scans of paleo
  and very small vertebrates. (Humphrey, Mirenkar)
• Huge variation in social and technical image and
  metadata acquisition

4
Image Lsid Resolution Servers

• Known
• UMB 13000 unedited images from skilled naturalist
• Metadata exif, taxonomy, habitat, location,
  voucher number for type specimen of identified
  taxon, part imaged. 13,000 Images in folders.
  Mine file and folder names and correlate to
  checklist(s) then metadata into MySQL with
  generated LSID.
• cf. ENBI report on Imaging Type Specimens
• Data ???
• Morphbank (U. Florida) from project to document
  morphological characters
• Metadata Darwin Core plus local attributes.
  Automated by Contribution process
• U.T. Austin X-Ray CT facility to document XRCT
• Metadata automated by scan configuration

5
Summary

• Resolution is easy.
• Acquiring metadata is hard.

6
UMB details

• Services implemented on sourceforge.lsid.net Java
  suite
• Authority, Data, and Metadata interfaces exposed
  as separate web services
• Omitted security service and assignment service
  (use adhoc assignment, not exposed. Would
  consider making assignment as part of the image
  deposit service).

7
Implementation issues

• Triples on-the-fly were too slow. We cache them
  in MySQL. Could use native triple store but
  havent yet encountered any use case except that
  needs it in the face of a shadow SQL metadata
  store and a warehouse model.
• Most integrated apps might be easier to do with
  something that appears to the outside like a
  triple store though.

8
Implementation issues

• Jena RDF serialization generated huge numbers of
  triples irrelevant to us (e.g. graph support).
  Result was intolerable performance so serialized
  with hibernate.org relational persistence
  framework. (Message from Mirenkar forcefully and
  us weakly there are no standards for serializing
  naturally occurring RDBs to RDF).

9
Warehousing vs distributed metadata stores

• Current resolution discovery scheme does not
  support multiple resolution services for a given
  LSID. Hence metadata cannot presently be
  distributed. Example distributed annotation.
  Bill may not have authority to add annotation to
  Susans metadata store but might still have
  valuable annotation which should be keyed by the
  LSID.

10
Warehousing vs distributed metadata

• Given some metadata values, how to find all the
  LSIDs that have that metadata value. Need entire
  metadata RDF store someplace (for each resolution
  service!) in order to make the query SELECT
  lsid WHERE metadataAttributeA(lsid) value_b
• Reasonable image RDF is 50-100 attributes.
  Reasonable personal image store is 105 images.
• This is not specific to RDF, but there is no
  history of supporting this kind of query at large
  scale.

11
Interesting research problem

• Typical utility in applications will(?) arise
  from metadata containing other LSIDs. But there
  are no standards for querying this or for
  recursive resolution. That is, the embedded LSID
  is a proxy for more metadata implied
  ontological relations. How to make resolvers
  accept ontological data, reason over it, and
  decide what recursive resolution should take
  place.

12
Grumble

• LSID Launchpad doesnt allow showing namespaces
  in the attribute-value pairs
• sourceforge.lsid.net framework does not support
  DDNS or some other magical multi-resolver
  discovery
• Jena rdf serialization doesnt seem to be
  scalable.