Integrate Content-Based Image Retrieval in Elasticsearch

Elastic Stack Elasticsearch
1

Hi,

I am wondering how dificult would be to integrate an API like LIRE (
http://www.semanticmetadata.net/lire/) into ElasticSearch. LIRe is an
Content-Based Image Retrieval Library based on Lucene. It relies on lucene
indexes to search image descriptors.
How the integration would be possible? Creating a new module in
ElasticSearch?

Regards,
Aécio Santos.
http://lapesi.ifpi.edu.br

2

If what it does is parse images and creates fields with possible more info on them, and possibly have specific set of queries associated with it, then it should not be too difficult to write a plugin that does that.
On Monday, January 24, 2011 at 9:25 PM, Aécio wrote:

Hi,

I am wondering how dificult would be to integrate an API like LIRE (Lire | SemanticMetadata.net) into Elasticsearch. LIRe is an Content-Based Image Retrieval Library based on Lucene. It relies on lucene indexes to search image descriptors.
How the integration would be possible? Creating a new module in Elasticsearch?

Regards,
Aécio Santos.
http://lapesi.ifpi.edu.br

3

What LIRe does is extract image features (from a
java.awt.image.BufferedImage) and store them as text in lucene documents
during indexing. This document can be indexed normally using Lucene's
IndexWriter. For search it receives a BufferedImage as query and returns
ImageSearchHits. It uses the IndexReader directly instead of IndexSearcher.
More info about LIRe can be found on this paper:
http://kusu.comp.nus.edu.sg/proceedings/mm08/mm/p1085.pdf

I don't know Elasticsearch well. There's any docs about how to create
plugins? How can I get started with Elasticsearch internals?

2011/1/24 Shay Banon shay.banon@elasticsearch.com

If what it does is parse images and creates fields with possible more
info on them, and possibly have specific set of queries associated with it,
then it should not be too difficult to write a plugin that does that.

On Monday, January 24, 2011 at 9:25 PM, Aécio wrote:

Hi,

I am wondering how dificult would be to integrate an API like LIRE (
Lire | SemanticMetadata.net) into Elasticsearch. LIRe is an
Content-Based Image Retrieval Library based on Lucene. It relies on lucene
indexes to search image descriptors.
How the integration would be possible? Creating a new module in
Elasticsearch?

Regards,
Aécio Santos.
http://lapesi.ifpi.edu.br

--
Regards,
Aécio Santos.
http://lapesi.ifpi.edu.br

4

On Tuesday, January 25, 2011 at 4:14 AM, Aécio wrote:

What LIRe does is extract image features (from a java.awt.image.BufferedImage) and store them as text in lucene documents during indexing. This document can be indexed normally using Lucene's IndexWriter. For search it receives a BufferedImage as query and returns ImageSearchHits. It uses the IndexReader directly instead of IndexSearcher.
More info about LIRe can be found on this paper: http://kusu.comp.nus.edu.sg/proceedings/mm08/mm/p1085.pdf

I don't know Elasticsearch well. There's any docs about how to create plugins? How can I get started with Elasticsearch internals?

Get the source code and have a look. Check the attachments plugin for introduction of a new field type.

2011/1/24 Shay Banon shay.banon@elasticsearch.com

If what it does is parse images and creates fields with possible more info on them, and possibly have specific set of queries associated with it, then it should not be too difficult to write a plugin that does that.
On Monday, January 24, 2011 at 9:25 PM, Aécio wrote:

Hi,

I am wondering how dificult would be to integrate an API like LIRE (Lire | SemanticMetadata.net) into Elasticsearch. LIRe is an Content-Based Image Retrieval Library based on Lucene. It relies on lucene indexes to search image descriptors.
How the integration would be possible? Creating a new module in Elasticsearch?

Regards,
Aécio Santos.
http://lapesi.ifpi.edu.br

--
Regards,
Aécio Santos.
http://lapesi.ifpi.edu.br

5

Hi Aécio,
I was wondering how far did you get into that project.
I am also interested in a solution combining both technologies. (Even a paid one)
Regards,
lemig

lemig at hotmail dot com