Image Painter: An Optimized Stroke-Based Algorithm for Artistic Image Stylization

This article introduces Image Painter (IP), a novel rule-based stroke-based rendering (SBR) algorithm that transforms photographs into painterly images using a sequence of brush strokes. IP employs a principled stroke initialization method combining connected-component labeling and principal component analysis, followed by a dual-stage optimization framework-forward optimization and backward optimization to refine strokes efficiently and interpretably. Experimental results demonstrate that IP outperforms state-of-the-art SBR methods across key metrics, including mean squared error, structural similarity index measure, peak signal-to-noise ratio, and learned perceptual image patch similarity, even with a limited number of strokes. IP supports diverse painting styles, such as oil sketch, watercolor, pastel, and spray painting, and has practical applications in digital art, step-by-step painting instruction, and synthetic dataset generation for learning-based SBR methods. These results highlight IPs effectiveness, flexibility, and potential for both creative and educational applications.