Artistic Portrait Drawing with Vector Strokes

TL;DR

Quantitative and qualitative evaluations both demonstrate that the portrait sketches generated by VectorPD can produce better visual effects than existing state-of-the-art methods, maintaining as much fidelity as possible at different levels of abstraction.

Abstract

In this paper, we present a method, VectorPD, for converting a given human face image into a vector portrait sketch. VectorPD supports different levels of abstraction by simply controlling the number of strokes. Since vector graphics are composed of different shape primitives, it is challenging for rendering complex faces to accurately express facial details and structure. To address this, VectorPD employs a novel two-round optimization mechanism. We first initialize the strokes with facial keypoints, and generate a basic portrait sketch by a CLIP-based Semantic Loss. Then we complete the face structure through VGG-based Structure Loss, and propose a novel Crop-based Shadow Loss to enrich the shadow details of the sketch, achieving a visually pleasing portrait sketch. Quantitative and qualitative evaluations both demonstrate that the portrait sketches generated by VectorPD can produce better visual effects than existing state-of-the-art methods, maintaining as much fidelity as possible at different levels of abstraction.

Figures (12)

• Figure 1: Examples of our method. VectorPD is able to generates vector portrait sketches at different levels of abstraction $L$, containing facial features and structure as much as possible. Left to right: The total number of strokes $S$ is set to 120, 80, 40 and the number of strokes in each optimization round is evenly divided.
• Figure 2: Editing the brush style on SVGs. Our method generates portrait sketches in vector form, which can be easily used by designers for further editing.
• Figure 3: An illustration of our framework. Given a target image $\mathcal{I}$, our method initiates by employing a facial extractor to derive essential facial keypoints within the facial region. These keypoints serve as the initial positions for strokes in the initial optimization round. The iterative optimization process is then guided by a CLIP-based Semantic Loss. Following the generation of the initial sketch results $\mathcal{S}_1$, we overlay the keypoints onto the facial contour obtained from contour extractor, initiating a second optimization round. The optimization continues iteratively, incorporating VGG-based Structure Loss and Crop-based Shadow Loss guidance until convergence is achieved, resulting in the final portrait sketch $\mathcal{S}_2$.
• Figure 4: More details on VectorPD. The top path represents the processing of the face extractor based on the user-defined number of facial strokes $N_f$. The path below shows the processing process of the contour extractor based on the user-defined number of contour strokes $N_c$. We employ different loss functions to guide the iterative optimization of strokes located at these key points.
• Figure 5: The optimization iteration process reflects two-round optimization mechanism of VectorPD.