LIRE Solr Integration Project

This is a Solr plugin for the LIRE content based image retrieval library, so basically it's for indexing images and then finding similar (looking) ones. The original library can be found at Github

The LIRE Solr plugin includes a RequestHandler for searching, an EntityProcessor for indexing, a ValueSource Parser for content based re-ranking and a parallel indexing application.

An outdated demo can be found at If you want to give it a try yourself, there is a docker image, which you use to run a pre-configured core on a Solr server. There's also a tool to create import XML files from Flickr. More information is available at src/main/docs/

If you need help on the plugin, please use the mailing list at lire-dev mailing list to ask questions. Additional documentation is available on src/main/docs/ If you need help with your project, please contact me, we also offer consulting services.

If you use LIRE Solr for scientific purposes, please cite the following paper:

Mathias Lux and Glenn Macstravic "The LIRE Request Handler: A Solr Plug-In for Large Scale Content Based Image Retrieval." MultiMedia Modeling. Springer International Publishing, 2014. Springer

The RequestHandler supports the following different types of queries

  1. Get random images ...
  2. Get images that are looking like the one with id ...
  3. Get images looking like the one found at url ...
  4. Get images with a feature vector like ...
  5. Extract histogram and hashes from an image URL ...


Supported values for feature field parameters, e.g. lireq?field=cl are ...

Also consult the Lire project and the documentation of features there. You can also extend the list of features by changing the FeatureRegistry in the LireSolr source.

The field parameter (partially) works with the LIRE request handler:

Getting random images

Returns randomly chosen images from the index. While it does not seem extremely helpful, it's actually great to find images to be used for example queries.


Search by ID

Returns images that look like the one with the given ID.


Search by URL

Returns images that look like the one found at the given URL.


Search by feature vector

Returns an image that looks like the one the given features were extracted. This method is used if the client extracts the features from the image, which makes sense if the image should not be submitted.




Extracting histograms

Extracts the histogram and the hashes of an image for use with the Lire sorting function. It will give you hashes and a truncated query for BitSampling (bs_list and bs_query) and MetricSpaces (ms_list and ms_query), but the latter only if it's available. the return values for bs_list and ms_list are ordered by ascending document frequency (BitSampling) and distance from the image to the respective reference point.

This can also be used to convert generic feature like the ones used for deep features, into base64 encoded feature strings and to obtain the appropriate queries for hashing based queries. If the field is sf_ha (or just sf), it is assumed that the extract parameter contains a comma separated list of doubles to be converted in a GenericGlobalShortFeatureVector.



Extraction from an image file:


results in

  "bs_list":["957","a26", "4a2", "276", ... ],
  "bs_query":"957 a26 4a2 276 e19 bf0 8b9 b2 ...",
  "ms_list":["R001902", "R001640", "R000511", ...],
  "ms_query":"R001902^1.00 R001640^0.88 ..."

Extraction from a double histogram:


results in

  "responseHeader": {
    "status": 0,
    "QTime": 2,
    "params": {
      "q": "*:*",
      "extract": "0,0,0,1,1,1,1",
      "field": "sf",
      "_": "1544015258504"
  "histogram": "AAAAAAAAAf8B\/wH\/Af8=",
  "bs_list": ["cd2", "2a5", "612", "d8", "510", "3e1", "d95", ...
  "bs_query": "cd2 2a5 612 d8 510 ..."

Function queries with lirefunc

The function lirefunc(arg1,arg2) is available for function queries. Two arguments are necessary and are defined as:

Note that if you send the parameters using an URL you might take extra care of the URL encoding, ie. white space, the "=" sign, etc.


If you extract the features yourself, use code like his one:

// ColorLayout
ColorLayout cl = new ColorLayout();
cl.extract( File("...")));
String arg1 = "cl";
String arg2 = Base64.getEncoder().encodeToString(cl.getByteArrayRepresentation());

PHOG ph = new PHOG();
ph.extract( File("...")));
String arg1 = "ph";
String arg2 = Base64.getEncoder().encodeToString(ph.getByteArrayRepresentation());

If you experiencing problems with a query having always the same results after changing the lirefunc parameters, you have to disable the cache of ordered search results by setting the size of the queryResultCacheto 0. The downside of this approach is that for paging the query has to be run through Solr over and over again.

<queryResultCache class="solr.LRUCache"


We assume you have a Solr server installed and running and you have already added a core. If not, check src/main/docs/ or don't even try but go for the docker image. First run the dist task by gradlew distForSolr command in folder where the build.gradle file is found to create a plugin jar. Then copy jars: cp ./dist/*.jar /opt/solr/server/solr-webapp/webapp/WEB-INF/lib/. Then add the new RequestHandler and the ValueSourceParser have to be registered in the solrconfig.xml file:

<requestHandler name="/lireq" class="net.semanticmetadata.lire.solr.LireRequestHandler">
    <lst name="defaults">
        <str name="echoParams">explicit</str>
        <str name="wt">json</str>
        <str name="indent">true</str>

<valueSourceParser name="lirefunc" 
    class="net.semanticmetadata.lire.solr.LireValueSourceParser" />

Use of the request handler is detailed above.

You'll also need the respective fields in the managed-schema file:

<!-- file path for ID, should be there already -->
<field name="id" type="string" indexed="true" stored="true" required="true" multiValued="false" />
<!-- the title of the image, e.g. the file name, optional -->
<field name="title" type="text_general" indexed="true" stored="true" multiValued="true"/>
<!-- the url where the image is to be downloaded, optional  -->
<field name="imgurl" type="string" indexed="true" stored="true" multiValued="false"/>
<!-- Dynamic fields for LIRE Solr -->
<dynamicField name="*_ha" type="text_ws" indexed="true" stored="false"/> <!-- if you are using BitSampling --> 
<dynamicField name="*_ms" type="text_ws" indexed="true" stored="false"/> <!-- if you are using Metric Spaces Indexing -->
<dynamicField name="*_hi" type="binaryDV" indexed="false" stored="true"/>

Do not forget to add the custom field at the very same file:

<fieldtype name="binaryDV" class="net.semanticmetadata.lire.solr.BinaryDocValuesField"/>


Check for indexing. It creates XML documents (either one per image or one single large file) to be sent to the Solr Server.


This help text is shown if you start the ParallelSolrIndexer with the '-h' option.

$> ParallelSolrIndexer -i <infile> [-o <outfile>] [-n <threads>] [-f] [-p] [-m <max_side_length>] [-r <full class name>] \\
         [-y <list of feature classes>]

Note: if you don't specify an outfile just ".xml" is appended to the input image for output. So there will be one XML file per image. Specifying an outfile will collect the information of all images in one single file.


The infile gives one image per line with the full path. You can create an infile easily on Windows with running in the parent directory of the images

$> dir /s /b *.jpg > infile.txt

On linux just use find, grep and whatever you find appropriate. With find it'd look like this assuming that you run it from the root directory:

$> find /[path-to-image-base-dir]/ -name *.jpg


The outfile from ParallelIndexer has to be send to the Solr server. Assuming the Solr server is local you may use

$> curl http://localhost:8983/solr/lire/update -H "Content-Type: text/xml" --data-binary "<delete><query>*:*</query></delete>"
$> curl http://localhost:8983/solr/lire/update -H "Content-Type: text/xml" --data-binary @outfile.xml
$> curl http://localhost:8983/solr/lire/update -H "Content-Type: text/xml" --data-binary "<commit/>"

You need to commit you changes! If your outfile exceeds 500MB, curl might complain. Then use split to cut it into pieces and repair the root tags (<add> and </add>). Here is an example how to do that with bash & linux (use Git Bash on Windows) under the assumption that the split leads to files {0, 1, 2, ..., n}

$> split -l 100000 -d images.xml images_
$> echo "</add>" >> images_00 
$> echo "</add>" >> images_01
$> echo "</add>" >> images_<n-1> 
$> sed -i.old '1s;^;<add>;' images_01
$> sed -i.old '1s;^;<add>;' images_02
$> sed -i.old '1s;^;<add>;' images_<n>

For small output files you may use the file upload option in the Solr admin interface.


Another way is to use the LireEntityProcessor. Then you have to reference the solr-data-config.xml file in the solrconfig.xml, and then give the configuration for the EntityProcessor like this:

    <dataSource name ="bin" type="BinFileDataSource" />
        <entity name="f"
                rootEntity="false" dataSource="null" onError="skip">
            <entity name="lire-test" processor="net.semanticmetadata.lire.solr.LireEntityProcessor" url="${f.fileAbsolutePath}" dataSource="bin"  onError="skip">
                <field column="id"/>
                <field column="cl_ha"/>
                <field column="cl_hi"/>
                <field column="ph_ha"/>
                <field column="ph_hi"/>
                <field column="oh_ha"/>
                <field column="oh_hi"/>
                <field column="jc_ha"/>
                <field column="jc_hi"/>
                <field column="eh_ha"/>
                <field column="eh_hi"/>

Mathias Lux, 2018-12-01