Abstract: An anti-aliased, or gray-edged, image is characterized by comparing the image to a number of loose-templates having a plurality of elements and at least one element having a range greater than zero. As families of loose-templates are compared to the anti-aliased input image, one or more sets of patterns, or features, are recognized and extracted to produce various screens of features that generally describe the anti-aliased input image. An arbitration technique is then performed on the sets of features to produce feature vectors which, in turn, can be used to control line-widths within the anti-aliased input image.