Image Classification Method Considering Overlapping and Correlation Between Probability Density Functions of Features

Image classification method considering overlapping and correlation between Probability Density Function: PDFs of features in the feature space in concern is proposed. As an example of the proposed method, sea ice type classification is demonstrated. A sea ice classification method based on eigen value decomposition with polarimetric SAR image data is proposed. Polarimetric SAR allows using surface and volume scattering characteristics of the targets such as odd/even/diffuse scattering, sphere/di-plane/helix scattering and so on. Based on such these eigen value decomposition, sea ice classification can be done with a variety of types of thin, rough and smooth surface sea ice. A comparative study among the proposed Maximum Likelihood based method with three components data of HH, VV and HV as well as odd/even/diffuse and sphere/di-plane/helix scattering components is conducted with PI-SAR of fully polarimetric SAR data of the Sea of Okhotsk area acquired on 23 Feb. 1999. It is found that the proposed method shows 1.3% improvement on percent correct classification compared to that of the existing method with three components and odd/even/diffuse scattering components.