Abstract:In robotic Chinese calligraphy, the brush stroke training models for Chinese hairy brushes play a crucial role in stroke generation. The method of combining end-to-end techniques and physical models requires further study, however, it is difficult to obtain large amounts of brush strokes for deep learning and training. To overcome this, we propose using a simulated brush model to train a generator based on the Bézier brush stroke model generator (B-BSMG), which was formed by two symmetric cubic Bézier curves according to the physical characteristics and writing posture of the brush. The B-BSMG can generate images for deep learning and training using a dataset simulated by the Bézier brush stroke model. Our renderer is based on parameterized brush strokes, providing a better foundation for deep learning or robotic writing. The results of several experiments prove that the proposed B-BSMG can generate stroke graphics well and outperforms other advanced stroke models.
What problem does this paper attempt to address?
### Problems Addressed by the Paper
The paper aims to address the key issue of stroke generation models in robotic calligraphy. Specifically, it focuses on how to generate high-quality stroke images by combining end-to-end techniques and physical models. However, existing methods face difficulties in acquiring a large amount of stroke data for deep learning and training. To solve this problem, the authors propose a generator based on a Bézier curve stroke model (B-BSMG), which trains the generator by simulating stroke data.
### Main Contributions
1. **Mathematical Definition of Simulated Stroke Model**: Establishes a mathematical relationship between the parameters of the simulated stroke model and the control parameters of the robotic arm.
2. **Construction of Simulated Stroke Model Generator**: Uses a differentiable neural network to construct the generator, improving flexibility and performance through optimized training.
3. **Experimental Validation**: Experimental results show that the proposed B-BSMG outperforms existing stroke-based neural renderers in generating stroke images.
### Method Overview
1. **Parameter Representation of Simulated Stroke Model**: By collecting real stroke data and locating four vertex coordinates, a multivariate linear regression method is used to analyze and establish the parameters of the simulated stroke model.
2. **Multivariate Linear Regression Model Parameters**: Defines a linear relationship between stroke parameters and robotic arm control parameters.
3. **Standard Stroke Design**: Designs cubic Bézier curves (CBC) to simulate the stroke patterns written by the robotic arm.
4. **B-BSMG Architecture**: Uses fully connected neural networks (FCN/MLP) and convolutional neural networks (CNN) to construct B-BSMG, with B-BSM parameters and starting coordinates as input, and stroke images as output.
5. **Loss Function**: Uses mean squared error (MSE) as the loss function to optimize all parameters through the backpropagation algorithm.
### Experiments and Analysis
1. **Experimental Hardware System**: Uses Aubo-I5 collaborative robotic arm, wolf hair brush, and Xuan paper for experiments.
2. **Data Collection**: Collects stroke images at different heights, tilt angles, and rotation angles by moving the robotic arm along the Z-axis until the brush tip just touches the paper.
3. **Stroke Contour Matching**: Normalizes, crops, binarizes, and contours the real stroke images, adjusting the simulated model parameters to match the real stroke images.
4. **Data Preprocessing**: Preprocesses the collected images, including resizing, cropping, and aligning, for comparison and similarity calculation.
5. **Training and Testing Dataset**: Generates multiple simulated B-BSM images as training and testing datasets.
6. **Training Process**: Updates network parameters using the MSE loss function and iteratively trains with the Adam algorithm.
7. **Results and Discussion**: Compared with classical models (such as Bézier curve model, circular model, and elliptical model), B-BSMG performs better morphologically and achieves the highest similarity of 94.59% in CSIM and SSIM metrics.
### Conclusion
The paper proposes a generator based on a Bézier curve stroke model (B-BSMG), successfully addressing the key issue of stroke generation models in robotic calligraphy. Experimental results show that B-BSMG has significant advantages in generating high-quality stroke images, providing new insights for the development of robotic calligraphy.