Abstract
This paper investigates the use of evolutionary algorithms (EAs) as a generative art tool, specifically focusing on the transition between two digital images. By employing a (1+1) EA with specialized mutation operators—including asymmetric, box, and strip mutations—the authors demonstrate that the intermediate evolutionary steps can produce unique aesthetic qualities, such as "glitch art" textures. The study utilizes images of Kyoto's Kinkaku-ji and Daigo-ji temples to showcase how these operators overcome the "coupon collector’s effect" to maintain visual dynamism throughout the transition process.
Introduction
While evolutionary algorithms have traditionally been used to optimize for a final output, their application in generative art often overlooks the aesthetic potential of the evolutionary path itself. This research shifts the focus from the destination to the journey, exploring how the stochastic transition from a starting image ($S$) to a target image ($T$) can be crafted into an artistic experience. The primary challenge is avoiding the "coupon collector’s effect," where progress slows significantly as the image nears the target. By introducing novel mutation operators, the authors create a visually engaging "glitch-inspired" trajectory.
Evolutionary Framework
The researchers employ a (1+1) Evolutionary Algorithm, a robust framework where a single parent generates one offspring through mutation. The offspring replaces the parent only if its fitness is equal to or greater than the parent's.
Fitness Function
Fitness is defined by the number of pixels in the current image that match the target image ($T$). This transforms the artistic transition into a variation of the classical OneMax problem. By framing the problem this way, the authors can leverage runtime analysis to ensure the transition remains computationally efficient and visually consistent.
Mutation Strategies
The core innovation lies in the mutation operators designed to produce specific textures and "glitches":
- Asymmetric Mutation: This operator assigns different flipping probabilities to pixels based on whether they currently match the start or target image. This ensures a constant rate of progress, achieving a runtime of $\Theta(n)$ and avoiding late-stage stagnation.
- Strip Mutation: This operator mutates horizontal or vertical strips of pixels (up to 180 pixels in length). This intentionally introduces linear artifacts reminiscent of digital "glitch art" or JPEG corruption.
- Box Mutation: By flipping $3 \times 3$ blocks of pixels, this operator creates a "chunky" or pixelated aesthetic, making changes more perceptible to the viewer.
Experimental Setup
The transition was tested using high-resolution images ($200 \times 200$ pixels) of two iconic Kyoto landmarks: the Kinkaku-ji Temple (The Golden Pavilion) as the starting point and the Daigo-ji Temple as the target.
Results and Discussion
The study's results are primarily qualitative, evaluating the visual appeal of the images at various stages of completion (e.g., 12.5% to 87.5%).
Operator Aesthetics
The choice of mutation operator significantly impacts the visual "feel" of the transition. Strip mutations create sharp, digital interruptions that mimic encoding errors. When combined with asymmetric mutation, the process produces a "blurred" or "old linen" effect, adding a layer of classical texture to the digital process.
The authors found that alternating between asymmetric pixel-wise mutation and structural mutations (strips/boxes) provided the most dynamic results. This combination ensures that the target image is revealed in a way that is both steady and surprising, maintaining viewer engagement throughout the process.
Conclusion
This research demonstrates that evolutionary algorithms can be more than just optimization tools; they can be instruments for generative art. By focusing on the transition process and designing operators that produce intentional artifacts, the authors have opened new avenues for digital aesthetics. Future work may involve more complex fitness functions based on structural similarity or interactive systems that allow artists to guide the "glitch" evolution in real-time.
Summary updated