Dynamic Line Drawing: A Motion Semantics-Based Photo-to-Sketch Rendering Algorithm

Generating line drawings from single photorealistic images featuring dynamic scenes or characters poses significant challenges. Existing algorithms predominantly focus on static scenes or objects, often producing uninspiring and monotonous line drawings when processing images with motion elements. This limitation primarily arises because these traditional methods emphasize static inherent features such as shape and three-dimensional geometry, while neglecting the motion semantics present in the original images, which are crucial for accurately depicting dynamics. To address these shortcomings, we introduce a novel algorithm for generating dynamic line drawings based on motion semantics. This algorithm incorporates a motion estimation network, a line drawing optimization algorithm, and a motion line generation algorithm. Through motion object mask estimation and motion compensation, our motion estimation network precisely predicts areas of motion and effectively removes motion blur from the previous frame. This ensures that subsequent line drawings accurately reflect the real motion of dynamic objects (foreground) in the scene and maintain rich line detail. Our line drawing optimization algorithm uses the motion mask to adjust the complexity and thickness of lines for static objects (background), thereby enhancing the depiction of dynamic foreground objects. Additionally, our motion line generation algorithm leverages a single static image and the estimated previous frame to calculate the direction of motion and interactively generate motion lines, enriching the dynamic texture of the line drawings. Experiments demonstrate that our method not only enhances the expressiveness of dynamic foreground objects in line drawings but also simplifies the background, directing viewers’ attention to the depicted motion.