PolSAR image analysis with Information Theory

• Grant number: 11/12386-5
• Support Opportunities: Research Grants - Visiting Researcher Grant - Brazil
• Start date: October 01, 2011
• End date: October 31, 2011
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Corina da Costa Freitas
• Grantee: Corina da Costa Freitas
• Visiting researcher: Alejandro Cesar Frery Orgambide
• Visiting researcher institution: Universidade Federal de Santa Maria (UFSM). Centro de Ciências Naturais e Exatas (CCNE), Brazil
• Host Institution: Instituto Nacional de Pesquisas Espaciais (INPE). Ministério da Ciência, Tecnologia e Inovação (Brasil). São José dos Campos , SP, Brazil
• Associated research grant: 08/58112-0 - Land use change in Amazonia: institutional analysis and modeling at multiple temporal and spatial scales, AP.PFPMCG.TEM

Abstract

This project aims at obtaining new results applying Information Theory tooks and techniques to the modelling of Polarimetric Synthetic Aperture Radar - PolSAR image analysis. These results will span theoretical advances, computational assessments and applications. They are related to deriving analytic expressions for entropies and relative entropies under the multilook complex multivariate Wishart distribution, and for distances betweeen pairs of this law and, possibly, other models for PolSAR data. The main challenge is related to the analytic difficulty of dealing with these models; they are defined in the space of complex Hermitian matrices and their transformations. Shannon, Rényi and Tsallis entropies will be studied, while distances will be derived under the h-phi divergence framework which includes the Kullback-Leibler, Rényi, Hellinger and chi-squared distances. Once an expression for an entropy is obtained, its ability to discriminate between different targets and its use as a test statistic will also be assessed. Distances members of the h-phi divergence family have known asymptotic properties which allows their use as test statatistics. The applicability of these new tests based on entropies and on distances to the problem of PolSAR image classification will be assessed. In particular, their small sample properties and numerical behavior with real data will be studied, in order to be able to use them in production image analysis systems. (AU)