JPEG-based image search broker

JPEG Broker demonstration

Access to the JPEG Broker demonstration

About DMAG (Distributed Multimedia Applications Group)

This software has been developed by the DMAG (Distributed Multimedia Applications Group), a research group of the Computer Architecture Dept. of the UPC-BarcelonaTECH (Universitat Politècnica de Catalunya).

Demo description

The aim of this prototype is to provide the ability to search images, from a central point, on different servers, simultaneously. Concretely, the image searcher consists of two modules, as sketched in figure 1: the core application, which implements the search engine, and the web portal, which is presented to the user as the front-end of the application.

Fig 1. System architecture

The search engine is based on the idea of query rewriting, that is done by a broker. Essentially, as it is shown in figure 2, the broker receives a query (QB) from the user and it rewrites it once for every metadata format that is supported (F1, F2, ..., Fn) via the corresponding interface (FB<->F1, FB<->F2, ..., FB<->Fn). The queries Q1, Q2, ..., Qn are generated and sent to each external system. The resulting data is processed back again by the corresponding interface, mapped back to the metadata format FB and presented to the user. Currently, the available servers (and their corresponding formats) are Flickr, Picasa and Panoramio.

Fig 2. Broker architecture

In order to ensure that the results are correct, they are reprocessed (or filtered) based on the search constraints, whenever it is possible. This filtering also expands the search capabilities of each server because it allows searches that would not be possible without doing it.

Moreover, the system is compliant with some JPEG standards (JPSearch), like the JPQF and the JPSearch Core used to accept/respond queries and to express the internal metadata model, respectively.

To sum up, an innovative image searcher that tries to solve some of the existing interoperability problem has been developed. It not only searches images on different servers using their public APIs, but also increases their search capabilities due to the reprocessing of results. Also, since the system is JPQF compliant, the search queries are highly customizable. Not only the search constraints can be defined, but also it is possible to specify how the results are returned and shown to the user. Finally, since some multi-threading and in-memory techniques have been applied, the searcher is reasonably efficient.

A more accurate explanation of this work can be found in the paper Image search based on a broker approach presented to the last Workshop on Interoperable Social Media Applications (WISMA 2010)