Goggles FLOSS

Introduction

Goggles FLOSS is a service that allow compare thousand of images very fast. This project is similar to Google Goggles but with FLOSS technologies. This projects uses the next FLOSS technologies:

• openCV: Use SURF extractors to obtain the interesting points of the images.
• Twisted: is an event-driven networking engine written in Python and licensed under the MIT license.
• FLANN: is a library for performing fast approximate nearest neighbor searches in high dimensional spaces. It contains a collection of algorithms we found to work best for nearest neighbor search and a system for automatically choosing the best algorithm and optimum parameters depending on the dataset.
• Lemur: The Lemur Project develops search engines, browser toolbars, text analysis tools, and data resources that support research and development of information retrieval and text mining software

Use Case: Android

Thanks to twisted, GogglesFLOSS can execute as web service and, for example, we can develope a simple application to send photos taken with the mobile and search in all images database. You can see this use case in the nexts videos:

Manual:

1. Start the server

rocapal#vega:~$ cd workspace/gogglesFLOSS/src
rocapal#vega:~/workspace/gogglesFLOSS/src$ twistd -n -y gogglesFLOSSd.py
Removing stale pidfile /home/rocapal/workspace/gogglesFLOSS/src/twistd.pid 
Initializing gogglesFLOSSd  ... 
[gogglesFLOSSd] Initializing SOAP service 
[gogglesFLOSSd] SOAP service initialized 
2010-10-23 18:59:19+0200 [-] Log opened. 
2010-10-23 18:59:19+0200 [-] twistd 10.1.0 (/usr/bin/python 2.6.6) starting up.
2010-10-23 18:59:19+0200 [-] reactor class: twisted.internet.selectreactor.SelectReactor. 
2010-10-23 18:59:19+0200 [-] twisted.web.server.Site starting on 9085
2010-10-23 18:59:19+0200 [-] Starting factory

2. Explore the client options:

 
rocapal#vega:~/workspace/gogglesFLOSS/client$ ./gogglesFLOSS-client.py 
Usage: ./gogglesFLOSS-client.py
Options: -a  	 add photos:
         -f  	 specified the JSON file
         -d  	 specified the images directory
         -q  	 send query file
         -g  	 generate the image index

3. Load images from directory

 
rocapal#vega:~/client$ ./gogglesFLOSS-client.py http:// 0.0.0.0:9085 -a -d /tmp/images/photos/
Sending photo 'c3938d4492935519d5a58238f805d58c[thumb].jpeg' ... 
Sending photo 'c3938d4492935519d5a58238f805d58c' ... 
[....]

4. Image query

First, it's neccesary build a JSON file to make the query. The JSON file is simple:

 
{ "command": { "id": "2", 
               "name": "query_photo", 
               "params": { 
                        "path_photo" : "/tmp/093e81245f99de65285df424bc153d87" 
                         }  
             }
}

And later, launch the query:

 
rocapal#vega:~/client$ ./gogglesFLOSS-client.py http:// 0.0.0.0:9085 -q query_photo.json

{ "code" : "200" , 
  "results": [{"id": "1057", "range" : "1550.25" },
              {"id": "1058", "range" : "562.689" },
              {"id": "1067", "range" : "311.051" },
              {"id": "1054", "range" : "310.265" },
              {"id": "1066", "range" : "280.508" },
              {"id": "1062", "range" : "250.856" },
              {"id": "1055", "range" : "245.079" },
              {"id": "1049", "range" : "239.380" },
              {"id": "1053", "range" : "239.357" },
              {"id": "1056", "range" : "217.369" }
} 

As you seen, the query returns a list of ten images more similar to query. The first field 'id' means the image identifier, remember that it's necessary that you assign an id to each image when you added it. So, you are the responsible to do the matching to obtain the image file. The second field, 'range', means the similitude in numeric terms, as a bigger number better results!.

Developers

If you can see or collaborate with the source code, you can download it from our git repository:

Git: http://git.libresoft.es/gogglesFLOSS/

Bugs & Problems

Lemur does't seem to work very well in some situations. If there are several images with a high different of keypoints number (for example image with 200 keypoints and other images with 3500 keypoints). It's possible that results may be incorrects.