Automatic mapping of woodlands in urbanized environments based on panchromatic IKONOS images - Case of the Montreal Metropolitan region

We present a performance analysis of second-order texture measures to automatically segment woodlands in a panchromatic IKONOS image. The methodology is divided in four steps: (1) local histogram equalization of the image, (2) calculation of the texture images according to the Unser's method, (3) supervised classification, and (4) post-classification filtering. The experimentation was carried out with an IKONOS image of the Montreal suburban area and shows that the four textures that optimize the woodland segmentation are the mean, contrast, correlation, and standard deviation, calculated over a window of dimension between 17 x 17 and 35 x 35 pixels, and with a sampling step of 1 or 2 pixels. The detection rate for woodlands of at least 1000 m(2) is high (around 90%) along with a low confusion with the other land cover classes (Kappa coefficients around 0.8). Although the study is based only on second-order texture measures, the results could serve as the basis of a more generic method for woodland inventory and monitoring in urban areas.