Correspondence analysis applied to textural features recognition

Abstract

Correspondence analysis (CA) is a powerful data analysis and decision support statistical method which provides information about the relative contribution of the different factors extracted from datasets under analysis. This method is used for dimensionality reduction and clustering interpretation in a wide range of applications. Our contribution highlights one of CA's potential application in the field of texture features extraction and classification in addition to demonstrating its capability of optimizing a nonlinear transformation of the grey level which may cause problems in other methods. A novel decision support image representation is introduced; its functionality is described and it is validated using nondestructive industrial inspection (NDII) and remote sensing satellite imagery. The behaviour of the new system is studied and its optimal parameters for texture recognition and dimensionality reduction are established by using factors analysis.