While the above algorithm just represents the colors of the image, for many styles it is also important to conserve salient image features such as edges or object boundaries. Hays and Essa [4] introduce a set of layers that refine a painterly representation at important places. Collomosse and Hall [5] use a mapping of strokes to geometric elements of different width according to the image salience. Both algorithms work well for NPR in general but are not easy to implement within our sequential feedback loop since they reorder strokes.