ScribGen

ScribGen: Generating Scribble Art Through Metaheuristics

Soumyaratna Debnath*, Ashish Tiwari, Shanmuganathan Raman

SIGGRAPH Asia 2024

Indian Institute Of Technology Gandhinagar, India
Paper        Demo       

Abstract

Scribble art, arising from chaos and randomness, remains one of the exceptionally attractive forms of art. Many works bridge the gap between sketches and images, but few translate images into meaningful chaotic expressions. While deep generative networks are known for understanding images, their ability to induce scribble drawings is under-explored. Unlike GAN-based approaches that generate line drawings, sketches, and contours, our work uses metaheuristics to produce scribble art from images. We extensively analyse various metaheuristic algorithms, demonstrating their optimal balance between creativity and computational efficiency. They offer better adaptability and accuracy than state-of-the-art deep generative models for image-to-scribble generation.

Contributions

A novel approach using metaheuristics to generate scribble art from images
Alternative to traditional deep learning based methods
No Need for supervised training
Analysis of various metaheuristic algorithm backbones for producing scribble art
Creates a balance between creativity and computation
Quantitively surpasses deep learning techniques in preserving visual similarity
Can serve as a data source for training deep learning models

ScribGen in Action

Image
Scrib-GA
Scrib-DE
Scrib-GSA
Scrib-PSO
Scrib-HHO
Image
Use the Slider to Interact
Image
Use the Slider to Interact
Image
Use the Slider to Interact
Image
Use the Slider to Interact
Image
Use the Slider to Interact

Citation

@incollection{debnath2024scribgen,
  title={ScribGen: Generating Scribble Art Through Metaheuristics},
  author={Debnath, Soumyaratna and Tiwari, Ashish and Raman, Shanmuganathan},
  booktitle={SIGGRAPH Asia 2024 Art Papers},
  pages={1--9},
  year={2024}
}