Animated Evolved Photomosaics: Constantly Becoming Other
Abstract.The interstices of art and technology have provided a space for creative practice, and also a nexus for research. This paper presents a project in which two artists worked collaboratively with two computer scientists using evolutionary programming to create photomosaic animations for large-screen projection in a gallery space. Constantly Becoming Other was created in 2005 as a way of connecting creative art practice and research with computer science work in the application of evolutionary algorithms to image generation. Our focus was on animating images made up of photomosaic tiles as the photomosaic image evolved from random towards the target. Our project, Constantly Becoming Other was part of a curated exhibition of computer art, entitled ArtEscapes 2007, Valencia, Spain.
1 Introduction
Art projects that cross over between art and technology have become increasingly common. Such art projects are also the focus of research whereby a synergy is created between artistic and scientific processes and methodologies. Prestigious academic conferences, such as ACM Multimedia and SIGGRAPH, provide examples of the benefits of interdisciplinary collaborations between researchers associated with art and technology. Working with artists and creative practitioners offers computer scientists unusual problems to investigate and different ways of approaching technical problems. The largely intuitive problem-solving processes of the artist are complementary to the often-systematic processes of a scientist. This is a generalisation that was pertinent to our collaboration and arises from the differences in the ways creative and scientific disciplines frame and value knowledge.
Biswas [1] conducted extensive research into the management of art-technology research teams and, based on his analysis, describes typical art-technology collaborations being driven by an artist's concept. An artist may dream up an image behaving in a specific way, either in terms of its animation, or in response to the movement of an external object, say a person; but the artist might not have the technical programming knowledge to make such an art object. The project will therefore remain an idea until the artist either learns computer programming or collaborates with a computer scientist. There are numerous collaborations of this nature. For example, Diana Domingues, an installation artist, and Eliseo Reatigui, a computer scientist, collaborated to make an interactive installation called I'Myth based on an algorithmic approach [2]. Ben Syverson, who is an artist with programming skills, explored how data sharing and distributed knowledge can be used for art through a system he calls likn [3]. Candy and Edmonds from the Creativity and Cognition Studios use four artist case studies to discuss the ways in which artists collaborate with technologists to create interactive art systems [4].
Evolved art is an intersection between creative arts practice and computer science, and it brings new research questions for both disciplines. The complexity and sophistication of evolutionary algorithms is a recognised area for computer science research. The technical complexity is informed by biological sciences, from which metaphors are drawn in order to describe mutation and crossover processes. The biological metaphors that underpin evolutionary programming can also inform creative arts practice. In turn, the questions and aesthetic outcomes envisaged by the artist can further inform the problem-solving inventiveness of a computer scientist. This resonates with postmodern preoccupations [5] in the creative arts disciplines, such as photography and video art.
2 The Collaboration
The collaboration between the four authors of this paper came about because two of the authors (a computer scientist and a multimedia designer) worked together in a cross-disciplinary undergraduate degree program at RMIT University, Melbourne. The computer scientist had been working in the area of evolutionary algorithms for image generation and showed some of the software associated with this work, which had already been programmed, to the multimedia designer. This led to the formation of a research group dedicated to exploring how evolutionary algorithms could be used for making art. Our collaborative process included weekly meetings with an open-ended agenda and flexible timeframe, during which we had numerous stimulating discussions. This ensured that communication was open and we had the opportunity to develop a strong understanding of the perspectives of our different disciplines.
We learned adapt to each other's thought patterns and languages through these weekly dialogues. Creative arts ways of knowing emphasise practice underpinned by philosophy and cultural theory as research. Computer scientists have a research practice founded on the principles of scientific experimental enquiry where results are measurable. We uncovered the implications of these differing approaches through first-hand experience. Our conversations revealed that there were different framing discourses and knowledge paradigms from which we approached our work. What seemed obvious for a computer scientist was not so for a humanities educated creative practitioner. Words like ‘subjective’ and ‘objective’ needed to be negotiated until a shared understanding was agreed. However, such a process is very time consuming and is dependent upon multidisciplinary researchers being open to each other's points of view.
The collaborations referred to by Biswas [1] were driven by the artists' visions, in as much as the software needed was developed in response to creative ideas and visualisations. Computer scientists would solve the problems generated by artistic goals. Our collaboration was a reversal of the typical ways in which art-technology collaborations work. The computer scientists had already developed a basic version of the photomosaic software and wanted to know, firstly, whether it was of interest to creative practitioners and, secondly, what they could do to improve it so that it would be of use to artists.
The artists in our team began by experimenting with the software, and then they drew up a wish list of things that needed further work. At first the software lacked an interface that made sense to the practitioners. Normally software used in creative arts has a graphical user interface. Computer scientists often work with command lines. This was the first item on the wish list. The programmers responded by developing a graphical user interface. The second item was a movie script that would make movies from a series of images. Our collaboration was an iterative cycle, whereby the programmers would listen to the artists, develop algorithms and scripts, and the artists would experiment with the software and return with feedback and further suggestions.
After the artists learned the software and what it could do, the team came up with a concept for an artwork that would explore the human face and the play of centripetal and centrifugal forces [6], and in which a face would emerge from scattered components only to fall apart again. The play of centripetal and centrifugal forces stand as a metaphor for people's sense of self being subject to constant transition and change, hence the title of our work: Constantly Becoming Other. We planned to show centripetal and centrifugal forces through an animation. The movie script allowed the artists to view animations of images runs relatively quickly, without having to use a commercial software package. At this point in the collaboration, the artists expressed a desire to be able to control the speed and smoothness of the motion in the animation. The computer scientists produced a script that allowed the artists to choose frames to be placed on the movie timeline.
There was an aspect of randomness that came from the behaviour of the evolutionary algorithm, in that the first generation was a random placement of photomosaic tiles on a grid that would gradually evolve over subsequent generations towards an image that was recognisable as a representation of a subject, for example, a portrait of a human face. The process is elaborated in Section 3 below. The artists conceptualised this evolutionary process as one in which coherence gradually emerged out of an initial random arrangement of mosaic tiles. The search and the fitness equation described in Section 3 thus became an integral part of the conceptual framework for the artwork. We experimented with various programming techniques to express the fluidity of identity and consciousness symbolised by a face and its desire for coherence. Discourses that informed and constituted our work concerned traces and memories, the use and reuse of images compiled from one viewpoint towards another viewpoint and the reordering and recomposing of the face. The image making technique we used is one in which concepts reinforce each other to make up a whole. The notions of randomness and chance in evolutionary programming stand for incoherence and tenuousness in life. We wanted the programming techniques to reinforce the metaphors of disintegration and reintegration of identity and selfhood. The movie script was modified accordingly.
Our objective was to create an artistic expression that could be exhibited in galleries. We chose the face as our subject matter in order to interrogate the flows of consciousness and how interactions and encounters shape our sense of self and other. Our aim was to produce an artifact that served as a commentary on portraiture as a dominant preoccupation in visual arts, using an evolutionary algorithm that evolved animated photomosaics. Initially, we wanted the animation to render in real time, whereby a photograph would be taken of the viewer and then that image would gradually emerge out of a set of miniature faces; however, this plan was put aside because the individual frames took too long to be rendered and we did not have the resources to develop a swifter algorithm. We compromised with a static movie that could be played in a loop.
3 Related Work
3.1 Portraits as Subjects and Objects
We had already chosen the human face as our subject matter, to express our concept of the fluidity of one's sense of self and of subjectivity. We turned to photography as a creative arts discipline to explore our concept further. Self-portraits and portraits reveal and document a human subject. The distinction between self-portraits and portraits is that in self-portraits the artist is also the subject, thus generating a doubled relation between the subject and object [7]. To explore this distinction further, we turn now to the semiotician Roland Barthes. In his seminal work on photography entitled Camera Lucida, Barthes reflects on this doubled relationship: ‘I am neither subject nor object but a subject who feels he is becoming an object…’ [8]. An artist who chooses to make a self-portrait engages in a process in which s/he is both the subject and object. This was the starting point in our concept development. The computer scientists and creative practitioners participated in this process together. We were all to be the subjects and objects of our artwork. We all took photographs of each other and all our faces were to appear in the artwork. These photographs were to be combined together later. The line between self-portraits and portraits was thus blurred.
We developed our concept further by locating our idea within contemporary cultural practices. Self-portraits are fetishised through popular culture and social software. Ascott [9] described ‘cyberception‘ as a perception of the self as a distributed mind and non-local body; and whereas the term is not in general usage, it foreshadowed the development of social software and our preoccupation with networked presences, including images of ourselves in sites such as Flickr, Youtube, MySpace and Facebook. Portraits and self-portraits are expressions of our desire to be present to ourselves and to be here, embodied in a tenuous and incoherent universe. According to Barthes, we read and make portraits as indexical signs, as indicators of the ‘real’ or the ‘authentic’ or the ‘that-has-been’[8]. This is an illusion by which we are willingly seduced. The face as a subject has a long history in both oriental and occidental art. The face is readily recognisable, no matter how distorted or how little information is on offer. McLeod [10] notes that a circle with two dots is interpreted as a face.
We wanted to position the human face as an iconic image, so as to create an open-ended visual text in order to extend the limits of possible viewer interpretations. An iconic image of the human face is one that focuses on the cardinal points – eyes and mouth – rather than on a recognisable image of a face. Icons in the Christian tradition have a history extending to Byzantium in AD332. The figures are non-naturalistic and appear to float in the background landscape. Perspectives are distorted and flattened. The key features of faces are emphasised [11]. Figure 1 below is a photograph of the religious icons used in St Isaac's cathedral in St Petersburg to depict Christian saints. To an Eastern Orthodox Christian, these icons and the saints depicted in them are instantly recognisable.
The aesthetic of iconic faces was one that appealed to all members of the interdisciplinary team. We were also inspired by the French conceptual artist Christian Boltanski's iconic memento mori Altar to the Lycee Chases [12], as well as by the Shroud of Turin [13]. Our aesthetic intention was to create iconic images of the face, referencing the Byzantine tradition [14] as well as more contemporary explorations of identity. These would then be animated, congealing time and furthering the illusion of ‘that-has-been.’ We decided to capture staged images of our faces – classic passport type shots against neutral backgrounds – for our source material.
We drew on Derrida's [15] notion that the subject ‘has no relation to him [/her]self that is not forced to defer itself by passing through the other in the form.of the eternal return.‘ Our faces would constantly be forced to defer, passing through various forms only to return over and over again as both photomosaic tiles and macro images. Video art is typically displayed in galleries in a looping format so that the narrative structure of a beginning, middle and end is disrupted; and thus the making of meaning is in perpetual deferral because viewers do not read the video in the same way as they read traditional cinema, where there is an obvious narrative resolution. We decided to work within the constraints of the video art format in our work Constantly Becoming Other, as it added another layer to our concept: that of time and constant motion, as well as metaphors of faces emerging out of incoherence only to disintegrate and remerge as other without clear narrative signposts of structure and temporal relations. Each face would be made up of miniatures of other faces, as a metaphor of the interdependence of individuals. We envisaged a work that would be a single channel loop of face animations materialising and dematerialising.
3.2 Mosaics, Photomosaics and Animated Photomosaics
Mosaics have a long tradition in art and allow artists to create images that may be viewed at both macro and micro levels. Viewers stand back to see the full picture, yet are enticed to come closer to explore the intricate detail and beauty of the individual mosaic tiles. Traditional mosaic images made up of many pieces of glass, stone or ceramic tiles may be found in many built environments, including Pompeii, Ravenna and Herculaneum in Italy and in shopping arcades in Melbourne, Australia. Figure 2 below is a photograph of mosaics in Pompeii taken by one of the authors.
The aesthetic appeal of mosaics is timeless. Mosaic work continues to be a technique used by fine artists such as Ellen Blakely, a San Francisco artist [16]. The development of the ability to digitise an image has now opened further possibilities for artists, permitting them to continue using traditional techniques in new ways; such as creating mosaics using electronic images.
Within the context of digital image processing, photographic mosaic work (and the term photographic mosaic) refers to the stitching together of photographic tiles to make up an overall image. Typically, photographic mosaics work on a micro and macro level of imagery. DominoPix, a computer graphics system attributed to Ken Knowlton [17], used the notion of decomposing a digital image into its constituent blocks. His system uses complete sets of dominoes as mosaic tiles to construct images within a digital environment. Silvers coined the term photomosaic to describe the results of his process of using digital images to make up a larger target image [18]. Silvers conceived of the idea of dividing a source (or target) image into regions and comparing them with source image portions, to determine the best available matching source image. Silvers' company, Runaway Technology, produces photographic mosaic stills. His portrait of Bill Gates [19] was exhibited in the National Portrait Gallery in London. Joan Fontcuberta's ‘Googlegrams’ are a further example of contemporary use of photographic mosaics. [20]. Photomosaic is thus a medium enabled by computer science available to artists.
Animations using photomosaics produce split-screen effects, whereby the final picture is revealed through a process of gradual disclosure. Such animations are already used for scientific data imaging, for example, NASA produced an animation visualising the Amazon rainforest using mosaic tiles representing data collected from satellites [21]. This technique is used to render images and image sequences that are meaningful to the viewer from data sets that could otherwise be overwhelming. The technique for animated photomosaics that aims gradually to disclose photomosaic frames towards the best approximation of the target image focuses on generating the best-fit frame at each iteration. It becomes an optimisation process. For example, Klein et al. [22] have developed a distance measure to assess based on average colour and three-dimensional wavelet decomposition signatures in the colour space. Their matching process also dynamically selects smaller tiling sequences from a collection of candidate source videos, and supports the use of different tiling shapes.
4 Generation of Animated Photomosaics Using Evolutionary Computing
Our animated photomosaic generation software uses an algorithm to generate photomosaic frames for animation purposes that is based on evolutionary computing or evolutionary computation. Evolutionary computing, a branch of artificial intelligence, is in turn based on Darwinian evolutionary theory. In Darwinian evolutionary theory, all life forms evolve over time through a developmental process popularly known as the ‘survival of the fittest.’ Our photomosaic algorithm uses an evolutionary computing technique known as the genetic algorithm, to evolve an original, randomly generated photomosaic into a ‘fitter’ photomosaic. (For simplicity, we will henceforth refer to this algorithm as the evolutionary algorithm.) Over time, the newly generated or evolved photomosaics converge towards a closer likeness of the target image of interest. The photomosaic animation algorithm generates a series of photographic images and then sequences selected images into a rendering of animated frames.
In this section of our paper we provide an introductory treatment of how the algorithm works. We begin by presenting some important definitions relevant to the evolutionary algorithm, and we explain how a single photomosaic is generated from fixed-sized, digital photographs denoted as tiles. We then describe the algorithm that creates the animated effect from a series of photomosaics, rendered into a frame-based movie that gradually evolves the photomosaics from a random grouping of tiles towards a close match of the intended target, and then back again to a random grouping.
4.1 Evolutionary Computing and Photomosaic Generation
Evolutionary computing is a branch of artificial intelligence that emulates Darwinian evolution to solve problems characterised by biota. Fogel [23] cites biological problems as being ‘typified by chaos, chance, temporality and non-linear intractivities’ and promotes evolutionary computing as a common way to solve difficult, real-world problems of a diverse nature.
In the context of photomosaic generation, our software, with its basis in evolutionary computing, treats a photomosaic as an individual to be evolved. The process of evolution moves from an arrangement of tiles, randomly placed on a fixed-size grid, towards an arrangement that best matches the known target image. This evolution occurs within a range of constraints imposed by the photomosaic generation software, henceforth referred to as MOSAIC.
To begin with, a set of tiles (or digital images) is made available to MOSAIC as its source of tiles for placement on the photomosaic grid. Tiles may be multiply placed on the grid, with the number of tiles made available typically being significantly less than the number of tiles needed to fill a grid (the photomosaic space). The choice of grid size is limited by screen size and resolution for movies, but this is not the case for still frames to be printed.
For illustrative purposes we use a small set of 7 tiles to explain the process of generating a single photomosaic, without loss of generality. MOSAIC numbers the tiles from 1 to 7, visualised as an array of tiles available for selection and use during photomosaic generation. As these tiles are digital images they may be replicated and re-used multiple times across the photomosaic space (the grid). Suppose our photomosaic has dimensions 4 rows by 6 columns. That is, it has slots or positions for up to 24 tiles to be placed in the space available. Initially, MOSAIC randomly selects tiles and places them on the grid (the photomosaic space). Here is a possible outcome of such placement, with the uniquely identified tiles placed on the 4 × 6 grid:
|
|
1 |
2 |
3 |
4 |
5 |
6 |
|
1 |
1 |
1 |
2 |
7 |
5 |
5 |
|
2 |
6 |
3 |
7 |
3 |
1 |
4 |
|
3 |
3 |
5 |
6 |
2 |
3 |
1 |
|
4 |
1 |
1 |
7 |
7 |
7 |
7 |
The photomosaic is generated by randomly selecting a tile and (randomly selecting) a grid position for the chosen tile. Initially, MOSAIC generates 120 separate grids. We now present the process that evolves these grids into a best-match final result: a photomosaic that matches most closely the original target image. The process uses an evolutionary computing solution.
4.2 Evolutionary Computation Operations: Elitism, Crossover and Mutation
By way of definition we denote the initial set of 120 grids as individuals, to use the terminology of evolutionary computing. These individuals represent a population, which, through a process of Darwinian evolution, will evolve into better or fitter individuals. Fitness is defined as follows: an individual X is fitter than some other individual Y if X has characteristics that are closer to a given target individual's characteristics than the characteristics of Y.
In the MOSAIC context an individual is a complete photomosaic, with all its N x M tiles, which may be viewed as being numbered from 1 to t. (In our mock example N=4 and M=6, so t is 24.) Thus, 120 complete photomosaics are initially generated (randomly) by distributing the 7 available tiles across the 24 grid positions or slots, in each of the 120 separate grids or individuals. Each such individual may be viewed as a linear array of tiles, exemplified by the following rendering of the above example matrix into its equivalent linear tile array.
|
1 |
1 |
2 |
7 |
5 |
5 |
6 |
3 |
7 |
3 |
1 |
4 |
3 |
5 |
6 |
2 |
3 |
1 |
1 |
1 |
7 |
7 |
7 |
7 |
The aim of MOSAIC may now be expressed as follows. Starting with a population of 120 randomly generated linear arrays of (N × M) numbers, apply an evolutionary algorithm [28] to this initial population to transform it into successive populations of fitter individuals. A transformed population results when the algorithm generates a new population from the previous one.
As each new population is generated, the best (fittest) individual is saved away, as a possible candidate for use in the photomosaic animation. We now present details of the algorithm used by MOSAIC, to generate each new population of (120) individuals. The algorithm employs elitism, crossovers and mutations to selected individuals in the current population, to form the new population.
Elitism is applied first and is set at 0.1, which means that 10% of the fittest individuals are promoted to the next generation. This ensures that we do not lose good individuals in evaluating the fittest ones. The remaining 90% (in the context of elitism set at 10%) are then subjected to crossover and mutation. Thus 108, of the 120 individuals are subjected to crossovers.
Crossover, set at 0.7, means that 70% of the 108 remaining individuals are paired up, randomly, two at a time, as a set of parents to be crossed over. Consider again the foregoing example individual (with tiles numbered from 1 to 24) and another individual. We denote these as P1 (parent 1) and P2 (parent 2).
|
P1 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
11 |
12 |
13 |
14 |
15 |
16 |
17 |
18 |
19 |
20 |
21 |
22 |
23 |
24 |
|
1 |
1 |
2 |
7 |
5 |
5 |
6 |
3 |
7 |
3 |
1 |
4 |
3 |
5 |
6 |
2 |
3 |
1 |
1 |
1 |
7 |
7 |
7 |
7 |
|
P2 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
11 |
12 |
13 |
14 |
15 |
16 |
17 |
18 |
19 |
20 |
21 |
22 |
23 |
24 |
|
2 |
5 |
5 |
6 |
4 |
3 |
2 |
5 |
1 |
2 |
2 |
4 |
4 |
1 |
6 |
7 |
6 |
5 |
4 |
3 |
2 |
1 |
5 |
6 |
For each couple tiles are swapped in two ways to create two children for the new generation. We employ two-point crossover. This means that two (like) random points in each individual (parent) are selected to generate a child. Let us suppose these two random points are denoted A and B. The child generated from the parents will inherit tiles from the beginning of the first parent up until point A, tiles from the second parent from A to B, and tiles from the first parent from point B to the last tile. The second child is generated by repeating this process, based on a new set of random points.
The process is best illustrated via the simpler single-point crossover, which means one random point in each individual is selected from both parents. In the example shown, tile number 8 from P1 and tile number 14 from P2 are selected. Information starting from the selected point onward will be swapped between the parents to create two new offspring, as shown below. The offspring are labelled P1′ and P2′. Note that crossover information from P1 and P2 is shown in red, that is, tiles 8 to 24 from P1 have been copied to positions 8 to 24 in P2, resulting in the offspring P2′, and tiles from positions 14 to 24 from P2 have been copied to positions 14 to 24 in P1, resulting in the offspring P1′.
|
P1′ |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
11 |
12 |
13 |
14 |
15 |
16 |
17 |
18 |
19 |
20 |
21 |
22 |
23 |
24 |
|
1 |
1 |
2 |
7 |
5 |
5 |
6 |
3 |
7 |
3 |
1 |
4 |
3 |
1 |
6 |
7 |
6 |
5 |
4 |
3 |
2 |
1 |
5 |
6 |
|
P2′ |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
10 |
11 |
12 |
13 |
14 |
15 |
16 |
17 |
18 |
19 |
20 |
21 |
22 |
23 |
24 |
|
2 |
5 |
5 |
6 |
4 |
3 |
2 |
3 |
7 |
3 |
1 |
4 |
3 |
5 |
6 |
2 |
3 |
1 |
1 |
1 |
7 |
7 |
7 |
7 |
These offspring become new individuals in the next generation of 108 (new) individuals. The crossover process is repeated for pairs of parents, to create two new offspring in each case.
Following crossover application, mutation is applied. When an individual is selected for mutation, tiles randomly selected from the tile repository are used to replace a corresponding number of tiles from the individual, located in randomly selected grid positions. In MOSAIC, mutation is set at 0.0001. This is used as a probability in a randomisation process that determines which tiles and how many will be replaced in the mutation process. The number of tiles replaced is limited to 9 in the present implementation of MOSAIC. Thus, 0 to 9 tiles will be replaced during the mutation phase.
In each new generation, the fitness algorithm is applied. The sum of differences of pixel grey-scale values in corresponding tiles, for each individual and the target, is computed. The best one is saved (for the animation). (MOSAIC works for coloured tiles and grid spaces as well, but discussion about colour is beyond the scope of this paper.) Elitism, crossover and mutation are then applied again to the new population; the process is repeated until some desired number of generations is reached or, typically, until the artist deems that a desired image is attained. In the latter case, the decision is subjective and dictated by aesthetic and conceptual choices. Heuristics are often deeply internalised and artists find it difficult to articulate these explicitly.
5 Making an Artwork with an Evolutionary Algorithm
Our objective was to animate photographic portrait images made up of photomosaic tiles rolling over as the mosaic image evolves from random towards the target. Animations are created by using some of the intermediate images created during the search for the best tiling as frames of a movie. figure 3 below shows a random distribution from the beginning of the evolutionary process. figure 4 shows the face emerging as the fitness algorithm selects best fits. figure 5 shows the final iteration, where the artists decided to halt the process. Figure 4 became the first and last frames in the movie timeline and figure 5 was the middle frame in the animation. Figure 2 was used twice. The tile set in Figures 4 to 6 shows miniature portrait photographs of the team.
The images for the tile set were made from the original portraits of the members of the research team. This meant that each portrait target could be made up of photomosaic tiles of the whole team's portraits. The images worked at both a macro and micro level, in line with the artistic concept goals of the project. The artists wanted to experiment with different tile sets to see the kinds of effects they produced.
Our approach to generating a photographic photomosaic requires a target image and a fixed set of tile images. Passport style portraits were used as target images, as contemporary iconic representations of the human face. To create a target that would facilitate a recognizable animation, the artists manipulated the images in image editing software to increase contrast and shadows to emphasise the eyes, nose and mouth in the portrait shots.
As described earlier, each time a new best image was found in an evolutionary run it was rendered and written as an image file. These images formed the basis for the individual frames of a movie. Some experimentation with parameters was needed to achieve the kind of motion aesthetic we were after. We were looking for an effect whereby it was apparent to the viewer that something was emerging, but for which the viewer had to wait to see what that would be. We wanted to create a hypnotic effect so that the viewer was somewhat mesmerised by the regular, rhythmic motion of the animation. We tried using every image created by the evolutionary algorithm, but we found that we wanted a faster rate of motion so that the rhythm would be apparent. Taking every kth image, k being between 1 and 10, worked well in most cases; with choice of k being very much dependent on the number of generations of best images created. To achieve the effect of coherence emerging from chaos, the face appearing from randomness, we copied the frames generated by the search algorithm, reversed the order, and constructed a movie file from the original and reversed frames. The movies clips were imported into commercial video editing software. Transitions in the form of cross-dissolves were added in order to create a ‘smooth’ effect. The completed work was exported as for output as a single channel DVD. Our final step was to submit a proposal that the work be included in a curated exhibition. Our proposal was successful. Our work was shown at Artescapes 2007 in Valencia, Spain.
6 Final Comments
We felt that the collaboration on which this research was based was both challenging and enriching for both the artists and computer scientists. It provided each of us with a window into the way knowledge is constructed and represented in a diametrically different discipline. In art, the intuitive and creative is privileged above procedural thought, whereas in computer science procedural logic is preferred. We built a machine using evolutionary computing algorithms capable of creating art by using a search-fitness function. The evolutionary operations, such as mutation and crossover, generated sequences that were converted to frames. In turn, the frames were used to create an animation as the final artifact. We chose to present the work as a static DVD; however, there are opportunities to develop this process further, so that a camera would photograph a viewer and the algorithm would run at a speed that would enable a real-time dynamic animation, composed of a predetermined set of tile images, in a gallery space. The viewer could also choose the tiles that would be used to make up his or her portrait. We have experimented with real-time in the form of a photo-booth trial, but more work needs to be done to improve run-time efficiency to achieve dynamism. This is an area for future computer science research.
The notion of locating our artifact within portraiture art was successful if one is to use acceptance into a curated show as a measure. In a gallery space video is typically played in loops, so that there is no clear beginning middle or end. The experience begins when the viewer first sees the work. The simplicity of Constantly Becoming Other was well suited to the gallery space and the viewer experience. The piece was viewed as designed.a large screen projection and was played through a DVD player.
Constantly Becoming Other showed us that the collaboration between artists and computer scientists was successful. Perhaps this can be explained in part by the artists. awareness of the implications of using photomosaic technology. They were prepared to work within the constraints of the technology in order to achieve an artwork. We have decided to continue that collaboration to produce new work through experimentation with evolutionary computing algorithms, to find new ways of working with target images. We are also exploring ways in which interactivity may be introduced, such that the viewer can interact with an installation to complete the artwork. Collaborations between artists and computer scientists push interdisciplinary boundaries and generate new fertile fields for research.
Acknowledgments
We thank Austin Wood for coding the original MOSAIC program; Irwan Hendra for subsequent modifications; Emil Mikulic and Peter Papagiannopoulos for technical assistance. Examples of our work may be accessed via: http://evol-art.cs.rmit.edu.au/
References
- [1]
- Biswas, Amitava. ‘Managing art-technology research collaborations.’ International Journal of Arts and Technology 1.1 (2008): 66–89.
- [2]
- Domingues, Diana and Reatigui, Eliseo. ‘Creating Synethetic Identities in I'Myth: Zapping Zone.’ Crossings: Electronic Journal of Art and Technology 5.1 (October 2007).
- [3]
- Syverson, Ben. ‘Meaning without Borders: likn and Distributed Knowledge.’ Leonardo Journal, 40.5 (2007). Available at: http://www.leonardo.info/isast/journal/toc405.html; accessed January 2012.
- [4]
- Candy, Linda and Edmonds, Ernest. ‘Interaction in Art and Technology.’ Crossings: Electronic Journal of Art and Technology 2.1 (March 2002).
- [5]
- Douglas, Crimp. ‘The Photographic Activity of Postmodernism.’ October, 15 (Winter 1980): 91–101.
- [6]
- Bakhtin, Mikhail. The Dialogic Imagination. Austin: University of Texas Press, 1981.
- [7]
- Jones, Amelia. ‘The "Eternal Return": Self-Portrait Photography as a Technology of Embodiment.’ Signs: Journal of Women in Culture and Society 27.4 (2002): 947–978.
- [8]
- Barthes, Roland. Camera Lucida: Reflections on Photography. New York: Wang and Hill, 1981; p.14.
- [9]
- Ascott, R. ‘Art Technology and Consciousness: The Technoetic paradigm.’ Convergence 4.3 (1998): 110–113.
- [10]
- McLeod, Scott. Understanding Comics. Northampton, Mass: Kitchen Sink Press and Paradox Press, 1993.
- [11]
- Cormack, R. Painting the Soul: Icons, Death Masks and Shrouds. London: Reaktion Books, 1997.
- [12]
- Boltanski, C. Altar to the Lycée Chases. Available at: http://www.moca.org/library/archive/exhibition/detail/1988/boltanski; accessed 06 October 2006.
- [13]
- Gove, H. Relic, icon, or hoax? Carbon dating the Turin shroud. Philadelphia: Institute of Physics Publishing, 1996.
- [14]
- Mango, C. The Art of the Byzantine Empire, 3121453: Sources and Documents. Englewood Cliffs, NJ: Prentice-Hall, 1972.
- [15]
- Derrida, Jacques, ‘Signature Event Context.’ In Margins of Philosophy. Chicago: University of Chicago Press, 1982/85; p.88.
- [16]
- Locktov, JoAnn and Plummer Clagett, Leslie. The Art Mosaic design: A Collections of Contemporary Artists. Massachusetts: Rockport Publishers, 1998.
- [17]
- Knowlton, K. DominoPix, US Patent No. 4,398,890, Representation of Designs. Available at: http://www.knowltonmosaics.com/ and http://www.metron.com/DominoPix/ (1983); accessed 02 November 2006.
- [18]
- Silvers, R. and Hawley, M. Photomosaics. New York: Henry Holt and Company, Inc., 1997.
- [19]
- Runaway Technology: Company web site: http://www.photomosaic.com/; accessed 06 October 2006.
- [20]
- Fontcuberta, Joan. Googlegrams. Selected Images, 31 January 2006–11 March 2006. Available at: http://www.zabriskiegallery.com/fontcuberta2006/images.htm; accessed 06 January 2012.
- [21]
- NASA. Mapping the Amazon: Mosaic tiles animation. Available at: http://svs.gsfc.nasa.gov/vis/a000000/a002400/a002405/; accessed 06 October 2006.
- [22]
- Klein, A., Grant, T., Finkelstein, A. and Cohen, M. ‘Video mosaics.’ Proceedings of the 2nd International Symposium on Non-photorealistic Animation and Rendering, Annency, France, 21.26 July 2002, pp 21–28.
- [23]
- Fogel, David B. ‘An introduction to evolutionary computation.’ In Evolutionary Computation: the fossil record, edited by David B. Fogel. Piscataway, NJ: IEEE Press, 1998; pp 23–41.