A texture based approach for ocean surface wind detection in SAR images

The radar backscattering on the sea surface is used for wind direction estimation. Current approaches at this problem are based on spectral methods which have limited spatial resolutions. Recent Image Processing methods use features extracted from local histograms of the image gradient direction. Here we report an alternative image processing approach which is based on the idea of extracting the preferred orientation of the textural patterns rather than the preferred orientation of the energy variation. Our method is based on a multichannel filtering approach which decomposes the input image over a set of matched filters. We also use a regularization operator for the detected orientations based on non linear diffusion. Experimental results are reported on synthetic images and real images.