EvoVec: Evolutionary Image Vectorization with Adaptive Curve Number and Color Gradients

Egor Bazhenov, Ivan Jarsky, Valeria Efimova, Sergey Muravyov

Abstract: Vector and raster graphics are the two main types of 2D images used in computer graphics. Raster graphics are images consisting of pixels (dots); vector images are created using mathematical objects such as lines, curves, and shapes. The main advantage of vector graphics is that they can be scaled without loss of quality, which is useful for advertising, design, frontend development, and other fields of application. At the moment, the issue of vectorization (conversion from raster to vector graphics) has not been fully resolved. There are two main approaches: deterministic algorithms and machine learning-based algorithms. Both of these types are not able to work with a color gradient and have other disadvantages, such as artifacts for deterministic algorithms, and extremely long working time and predefined curve number for machine learning-based algorithms. To solve the problems of existing solutions, we propose an evolutionary algorithm for image vectorization. Its main idea is to iteratively improve vector images using mutations and crossover. The proposed algorithm does not require any necessary parameters other than the original image and can process color gradients. The results of comparison with existing solutions show that our algorithm qualitatively and quickly vectorize images. Particularly, our approach outperforms others in terms of pixel-by-pixel MSE by \(15\%\). The implementation is publicly available (https://github.com/EgorBa/EvoVec-Evolutionary-Image-Vectorization).

External IDs:dblp:conf/ppsn/BazhenovJEM24