Skip to content

Contains a set of analysis processes on streams of incoming media items.

License

Notifications You must be signed in to change notification settings

chazelton331/socialsensor-multimedia-analysis

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

48 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

socialsensor-multimedia-analysis

Contains a set of analysis processes on streams of incoming media items.

Counters

Counters is a set of services that executed periodically by extending the class TimeTask of Java. In the package eu.socialsensor.framework.multimedia.counters there are four counters for counting contributors, domains and tags by submiting map/reduce jobs in mongodb.

ContributorsCounter

ContributorsCounter is based on the MapReduce operation of mongodb and counts the number of unique contributors (users) per timeslot.

To initialize and execute contributors counter every 30 minutes run the following code:

  ContributorsCounter counter = new ContributorsCounter("hostname", "dbname", "collection", "output");
  Timer call = new Timer();
  call.scheduleAtFixedRate(counter, 0, 30*60*1000);

The service submit the the following map/reduce javascript functions and mongodb engine executes them and writes the results in the collection "output"

map function

  function() {
  	var k = {stream:this.streamId, author:this.author}; 
	emit(k, 1);
  }

reduce function

function(previous, current) {  
    	var count = 0;
    	for (index in current) {
    		count += current[index];
    	}
    	return count;
    }

TagsCounter

The tags counter is initialized and executed in a similar manner.

map function

function() {
    	for (index in this.tags) {
    		var tag = this.tags[index]; 
    		var tmp = \"=\";
    		if(tag.length<20 && tag.indexOf(tmp) == -1) {
    			emit(tag.toLowerCase(), 1);
    		}
    	}
    }

DomainsCounter

The domains counter is initialized and executed in a similar manner.

map function

function() {
	var domain = this.url.match(/:\\/\\/(.[^/]+)/)[1];
	emit(domain, 1);
    }

Geospatial Analysis

The package eu.socialsensor.framework.multimedia.spatial contains a set of classes for the periodic clustering of geo-tagged media items based on their spatial proximity. The proximity of media items is based om the Vincenty geiodesic distance between two geo-points. The used clustering algorithm is BIRCH. The resulted clusters indexed in a solr collection.

To initialize and execute birch clustering:

Timer timer = new Timer(); 
timer.schedule(
	new BIRCHClusterer(
		"mongoHostname", 
		"mongoDBname"
		"collection", 
		"solrHostname"), 
	0, 
	15*60*1000
);

Visual Analysis

The package eu.socialsensor.framework.multimedia.visual contains a set of classes for the periodic clustering of media items based on their visual similarity. The visual similarity of media items is based on a combination of SURF descriptors aggregated with the VLAD scheme. For the extraction of VLAD+SURF descriptors is used the implemetation provided by Socialsensor: multimedia-indexing For the clustering, SCAN graph clustering algorithm is used.

Additional information

###Project dependencies### The computational-verification project is dependent on two SocialSensor projects:

###Contact information### For further details, contact Symeon Papadopoulos ([email protected]) or Manos Schinas ([email protected]).

About

Contains a set of analysis processes on streams of incoming media items.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages