A new dimensionality reduction method for seabed characterization: supervised Curvilinear Component Analysis

Abstract

In this paper, we present a new method for dimensionality reduction, called supervised Curvilinear Component Analysis, for the classification of sonar images task using support vector machines. Indeed it is important in many underwater applications to get tools that give automatically the kind of sediments. This method derives from the known method Curvilinear Component Analysis. It gives good results for data not highly overlapped. We have used this method after a feature extraction step based on wavelet decomposition applied to our sonar images database.