Synopsis:
Similarity search is required in many areas. For example, music matching and binary program matching require a similarity search engine. Nowadays, it is common to hear news of projects like "photosynth" that heavily rely on similarity search. OBSearch is a similarity search engine that can help you to create an interesting and new application!
Features:
- Single-computer or distributed mode. (1)
- Designed to handle efficiently heavy objects (trees, graphs).
- Code generation for each primitive datatype (byte, short, int, long, float, double) for maximum performance.
- The API is compact and easy to understand.
- Stability and scalability: OBSearch's secondary storage backend is Oracle's Berkeley DB. An extensive test suite makes sure that data integrity is preserved.
- Cutting edge: We strive to put together the latest algorithms the scientific community has to offer. For example, we use the K-means++ algorithm in one of our indexes.
- OBSearch can live on top of any B-Tree (or anything that allows range queries for one dimension). For example, Amazon's cloud service SimpleDB http://www.amazon.com/gp/browse.html?node=342335011
OBSearch's homepage is: http://obsearch.berlios.de/
This project started as part of Google Summer of Code 2007. The mentoring organization was Portland State University.
Technicalities:
Available indexes:
- IDistance
- Extended Pyramid Technique
- P+Tree
- D-Index (with rho=0)
- K-nn graph
(1) The first version 0.7-GSOC supports distributed access by JXTA. Right now, we are evaluating the distributed technology we will employ in the future.
Mailing lists: Announce: http://groups.google.com/group/obsearch-announce Users: http://groups.google.com/group/obsearch-users Dev: http://groups.google.com/group/obsearch-developers
