SBART 2.4: An IEC Tool for Creating Two-Dimensional Images, Movies and Collages

Abstract

In this article, the author gives an overview of SBART 2.4, an interactive system used to create abstract two-dimensional images, collages and movies. The system, one of the successors of Karl Sims's system, runs on a small computer that uses a function to calculate the color value of each pixel as a genotype. All of the ranges and domains are three-dimensional vectors. The system utilizes a multi-field user interface to enhance the diversity of production and has optional facilities that allow the creation of collages of external images or short movies.

In 1991, Karl Sims developed a system that enables users to create abstract drawings merely by selecting their favorite images from a computer screen [1]. It was very innovative and great creative impetus to both artists and computer scientists. I was one of computer scientists who planned to implement a similar system to explore the further possibilities of this new technology [2]. I began developing SBART (Simulated Breeding for ART) in 1993 as a successor to Karl Sims's system, with certain unique extensions. Such systems have become recognized as artistic applications of interactive evolutionary computing (IEC) [3]. As in other IEC systems, SBART's main mechanisms are an iteration of visualization, selection by the user and genetic operation. The window, or "field window," that appears on the screen at the beginning consists of 20 sub-windows, which display individual images of the current population (see Color Plate A No. 3) [4].

The following sections describe the distinctive features of SBART 2.4, including genotype representation, multi-field user interface and facilities for creating collages and short movies.

Genotype

An essential feature of evolutionary computing is the automatic search for better solutions by selection and modification. Each candidate is called "individual" that consists of "genotype" and "phenotype" following biological terminology. Genotype is the genetic information inherited from an original candidate, or a parent, to a modified candidate, or a child. Phenotype is the functional feature produced from genotype that is the object for evaluation and selection. A chromosome, or data unit containing the genotype, is a tree structure that represents a mathematical formula, as in genetic programming [5], where the non-terminal points are a number of numerical functions and the terminal symbols are three-dimensional vectors of constants and variables. A constant is a triplet of unique scalar values. For instance, in the set of coordinates [-1,1], these values would be (0.2, 0.2, 0.2). A variable vector is a permutation of x, y and 0. For example, x, y, 0), (x, 0, y), (y, x, 0) and so on, with six types in total. When creating a movie, each 0 in these variable vectors is replaced with the time variable t. In contrast with Sims's system, the domain and range of calculation are 3D vectors. Some functions, such as addition, subtraction and other ordinary numerical functions, calculate three values independently. In the case of addition, for example:

The other exceptional functions calculate their values by combining given values. For example, the function "max" takes two arguments, compares their first elements, and then returns any argument in which the first element is larger than the other one. The definition of "max" is:

The final resulting value of a calculation is mapped onto a point in the hue, saturation and brightness (HSB) color space,

Fig. 1..

SBART screenshot, 2001. (© Tatsuo Unemi) Dragging and dropping an individual for migration.

Fig. 2..

SBART Illustrator drawing, 2001. (© Tatsuo Unemi) Blend ratio for z in the cases of one, two, three and four images, from top to bottom.

substituting the coordinate of the pixel with the variables x and y. If the resulting value of any element is beyond the border of the range [-1,1], it is forced into the range by a saw-shaped function as follows:

where k(x) = |(x+1)/2|.

The value is finally mapped onto the RGB color space to paint each pixel...