6. Semantics
Both introduced algorithms handle the whole picture uniformly. Each painting technique described in Section 4 generates strokes that only depend on the brightness and gradients. There is no differentiation between objects. A common technique in the art of painting is to change the painting style according to objects. Background is painted with less detail, while the foreground is painted with thinner strokes to enhance the details and to let the foreground appear sharper [18]. Humans recognize semantics of objects and adapt their painting style. Grass for example is painted with short thin strokes, which are aligned perpendicular to the ground. Skin is represented by strokes, which are painted into one another to construct a uniform area. There are only a few painterly rendering techniques that take semantics and the differentiation of objects into account. Most of the algorithms like Hertzmann's painterly rendering method [3] just use region information to choose the size of the brushes used. Since the use of the additional information about objects could improve painterly rendering results, this suggests a fruitful direction for future work, which would be to explore painting different styles.