CompEOD Executive Summary

Project Title:

Compression based Analysis for Information Retrieval from Earth Observation Databases (CompEOD)

 

Contracting Authority: 

Unitatea Executiva pentru Finantarea Invatamantului Superior, a Cercetarii, Dezvoltarii si Inovarii (UEFISCDI)

 

Program:

Resurse Umane, Proiecte de cercetarepostdoctorala - tip PD

 

Project Code: 

PN-II-RU-PD-2012-3-0477

 

Contract Number:

84 / 30.04.2013

 

Executive Summary

This project aims at investigating and combining compression techniques due to their ability to yield comparable performance independently from the parameters extraction and modification steps. We propose to create an unsupervised procedure, parameter free that will generalize the issue of information extraction for all the type of analyzed data. The results will include information related to all the levels of classic content representation. Moreover, this enables the development of information retrieval methodologies with a minimalist approach, greatly simplifying its implementation and usage. By compression, image content will be concentrated to a minimum size, encapsulating at the same time all the distinctive features, with local or general character. The semantic meaning of the objects is preserved in latent manner since compression based methods are parameter free. Features extraction and several analysis levels are no longer necessary, reducing the procedure complexity and computational time. Through compression, a single vector is generated and can be used as a single feature comprising all the objects’ characteristics. Then, the features are clustered using compression based similarity measures to produce the information content indexes. The proposed procedure will be applied on Earth Observation (EO) databases, and its generality will be proven by testing several types of data. 

 

Objectives

 

1. Investigate and combine compression techniques based on their ability to yield comparable performance independently from the parameters extraction and modification steps.

 

2. Create an unsupervised procedure, parameter free that will generalize the issue of information extraction for all the type of analyzed data. The results will include information related to all the levels of classic content representation. The subjective information removal means that user subjective choices will not bias the process, avoiding risks such as failure at finding meaningful patterns because of poorly chosen parameter settings, incorrect discovery of patterns which do not exist, or overestimation of the importance of a parameter.

 

3. Validation and evaluation on Earth Observation datasets.

 

Original contributions

 

Theoretical aspect: The project involves the study of the dictionary generation, data compression algorithms and compression based similarity measures in order to reduce the storage, the computational speed and the information lost. The theoretical approach will result in a procedure for indexing and information retrieval with a very strong practical character.

 

Practical aspect: The generality of the method will allow its use in a wide range of application fields, such as text, multimedia and EO databases. The project will focus on the analysis of large EO image collections.