In [45], Haeberli describes a technique for using filters in the form of brush strokes to create abstract images (impressionistic paintings) from source images. The output image is generated by rendering an ordered list of brush strokes. Each brush stroke contains color, shape, size, and orientation information. Typically the color information is determined by sampling the corresponding location in the source image. The size, shape, and orientation information are generated from user input in an interactive painting application. The paper also describes some novel algorithms for generating brush stroke geometry. One example is the use of a depth buffered cone with the base of the cone parallel to the image plane at each stroke location. This algorithm results in a series of Dirichlet domains [83] where the color of each domain is sampled from the source image.
Additional effects can be achieved by preprocessing the input image. For example, the contrast can be enhanced, or the image sharpened using simple image processing techniques. Edge detection operators can be used to recover paths for brush strokes to follow. These operations can be automated and combined with stochastic methods to choose brush shape and size to generate brush strokes automatically.