Texture Orientation and Period Estimator for Discriminating Between Forests, Orchards, Vineyards, and Tilled Fields

Current orientation estimation methods are ill adapted to arbitrarily shaped regions and do not handle multichannel images optimally. Fourier-based ones have trouble with texton-based textures. The authors present a novel orientation and period estimator suitable for texton-based multichannel arbitrarily shaped texture patches with one or two orientations. They successfully apply it to the problem of discriminating forests, orchards, vineyards, tilled fields, and untilled fields in high-resolution aerial images, obtaining 95.5% accuracy; 82% of estimated directions are within 3deg of their correct value, and 81% of estimated periods are within one pixel of their correct value