SmartTracer: Interactive Tracing-based Stroke Extraction for Complex Line Art

Abstract

Strokes are fundamental for understanding and editing drawings, as they constitute the basic units artists use to create line art. However, because different users may interpret stroke segmentation patterns differently within the same high-quality line drawing with complex topology, existing methods cannot efficiently support interactive editing that aligns with user intent. In this paper, we address the task of stroke segmentation, aiming to generate masks for natural and high-quality strokes from raster line drawings with complex topologies. Given a stroke prompt, such as a point or a short mark, our method employs convolutional neural networks (CNNs) to predict the corresponding stroke with reasonable connectivity. Building on this single-stroke extraction model, we further propose a graph-based algorithm to segment all strokes in a drawing while preserving user control in topologically complex regions. We also develop an interactive system, SmartTracer, to demonstrate the effectiveness of both our single-stroke and multi-stroke segmentation methods. Quantitative comparisons and user studies show that SmartTracer achieves superior stroke segmentation quality compared with existing methods, while also providing more convenient and efficient stroke extraction operations than commercial software.