Pixel Classifier Algorithm Overview.

During training, grids were extracted from the expert labeled training windows. The input patterns determined the size, scaling, and quantization of these grids. Once grids were extracted, hash codes were computed to reduce dimensionality of the input space. Counts were then stored in a database for both positive and negative examples. During testing, the same process occurred for an unknown image window. Once a hash code was computed for a pixel in the unknown image window, it was looked up in the database of counts to produce a classification. See text for details.