TL;DR
This paper presents a novel image-based technique for creating realistic portrait engravings by combining ordered dithering with a simple 3D head model. The method warps the dither matrix to make engraving lines follow the contours of the face, enhancing the sense of form and texture, and also demonstrates how to extend this to color engravings.
What Problem Does This Paper Address?
Creating realistic artistic renderings of images, particularly portraits, is a long-standing challenge in computer graphics. While traditional engraving techniques rely on skilled artists to imbue lines with form and texture, automating this process while maintaining artistic quality is difficult. Existing image-based engraving methods often require significant user input or complex 3D models. This paper aims to bridge this gap by developing a simpler, image-driven approach that can generate convincing engraved portraits, capturing the nuances of facial structure without requiring a full 3D scan.
Methods
The core of the method involves two main components: ordered dithering and a proxy 3D head geometry.
- Ordered Dithering: The paper first constructs a specialized dither matrix designed to produce hatching and cross-hatching patterns. This matrix is carefully crafted to create lines of varying thickness and density, mimicking traditional engraving. The authors describe a process for combining horizontal and vertical line patterns, with parameters to control the prominence of cross-hatching.
- Proxy 3D Head Geometry: To make the engraving lines follow the face's curvature, a simplified 3D geometry is estimated. This is achieved by first detecting facial landmarks using a tool like OpenFace. From these landmarks, a rough cylindrical model of the head is constructed. This proxy geometry is then used to warp the dither matrix. Essentially, the dither matrix is deformed so that when applied to the image, the resulting engraving lines curve around the estimated shape of the face, creating a more natural and stylized appearance.
- Color Engraving: The authors extend their method to color by exploring two approaches: masking a color image with a black-and-white engraving, and applying the dither matrix independently to each RGB channel with potential shifts to create vibrant color separation effects.
Datasets
Not discussed in the paper. The method is applied to input portrait images.
Key Insights and Ideas
- Warping Dither Matrices for 3D Effect: The most significant insight is the clever use of a simple 3D proxy geometry to deform the dither matrix. This allows the 2D dithering process to simulate the curvature of a 3D object, a crucial element for realistic portrait engraving.
- Proxy Geometry Sufficiency: The paper demonstrates that a very simple, rough 3D model (like a cylinder) is sufficient to achieve a compelling stylistic effect, avoiding the need for complex and potentially error-prone full 3D reconstruction.
- Controllable Stylization: The method offers control over key engraving parameters, such as the density of lines and the amount of cross-hatching, allowing for a range of artistic styles.
- Color Extension: The exploration of color engraving, particularly the channel-shifted approach, shows a promising direction for creating visually rich and dynamic engraved outputs.
Key Results
The paper demonstrates the effectiveness of its method through qualitative results. The generated engravings exhibit curved lines that follow the facial contours, enhancing the perceived depth and form of the portraits. The authors show examples of both black-and-white and color engravings, highlighting the versatility of their approach. They also show how variations in parameters (like the scaling factor 'S' for cross-hatching or dither matrix size) affect the final output, allowing for different engraving densities and styles.
Limitations and Future Work
- Limitations:
- The method relies on accurate facial landmark detection. Errors in detection could lead to mismatches in the engraving line orientation.
- The current approach is primarily demonstrated on frontal, vertically aligned faces. Handling extreme poses or non-frontal views might require more sophisticated 3D estimation or warping strategies.
- The paper notes that using a simple cylinder as proxy geometry can lead to artifacts, especially at the top of the head or where gradients are high, which are partially addressed by blurring.
- Future Work:
- The authors suggest exploring more sophisticated 3D proxy geometries or using depth estimation methods (as briefly shown with a CNN-based approach) for potentially more accurate results.
- Developing a more robust method for handling non-frontal poses and varying head orientations is a natural extension.
- Further investigation into color engraving techniques to achieve even richer and more artistic results is also implied.
Summary updated