An unsupervised distance learning framework for multimedia retrieval

Abstract

Due to the increasing availability of image and multimedia collections, unsupervised post-processing methods, which are capable of improving the effectiveness of retrieval results without the need of user intervention, have become indispensable. This paper presents the Unsupervised Distance Learning Framework (UDLF), a software which enables an easy use and evaluation of unsupervised learning methods. The framework defines a broad model, allowing the implementation of different unsupervised methods and supporting diverse file formats for input and output. Seven different unsupervised methods are initially available in the framework. Executions and experiments can be easily defined by setting a configuration file. The framework also includes the evaluation of the retrieval results exporting visual output results, computing effectiveness and efficiency measures. The source-code is public available, such that anyone can freely access, use, change, and share the software under the terms of the GPLv2 license.