Transformation into ondolettes for extraction of texture-color. Application to the combined classification of images (HRV) of SPOT

Texture and colour are often used in a separated scheme without consideration of their mutual interactions. In this work they are combined in an integrative approach using the wavelet transformation technique. For that, texture-colour features are defined as wavelet coefficients covariance after colour space transformation. Colour-texture new channels are then used in addition to the radiometric channels to conduct a supervised classification process. In order to asses the texture-colour contribution, two classifications have been performed on the HRV (XS) SPOT images over the city of Oran (west of Algeria) taken as a test zone for its thematic variety, heterogeneity and relatively uneven topography. The first approach uses only spectral information while the second combines the radiometric and texture-colour information in a multi-resolution classification scheme. Results show that the proposed approach brings significant improvement of 18.9% to the global classification rate compared to 75.9% given by the spectral approach. The texture-colour feature allowed better discrimination of confused items.