Transcript Depixelizing Pixel Art
Depixelizing Pixel Art
像素风格画的矢量化 主讲:钱博磊 2012-01-04
•
研究对象
矢量图 -轮廓 轮廓所封闭区域的颜色 像素风格画 图像特征为像素级别
•
实现步骤
1. 图像重构 reshape the pixel cells 2. 轮廓提取及拟合 edge detecting and fitting curves 3. 图像绘制 rendering
•
图像重构
1.1 重构像素单元格 reshape pixel cell 目标:解决像素图连接模糊性问题,使有 相同颜色与相同特 征 的pixel至少拥有一条共享边
•
图像重构
1.1.1 构造相似图 similarity graph
•
图像重构
1.1.1 构造相似图 similarity graph curves sparse pixels islands
•
图像重构
1.1.2 调整像素单元格 voronoi diagram
•
实现步骤
1. 图像重构 reshape the pixel cells 2. 轮廓提取及拟合 edge detecting and fitting curves
•
轮廓提取及拟合
2.1 edge detecting and fitting T-junction 问题 • • 连接仅包含vanlence-2结点的可见边 用quadratic B-spline curve拟合 shading edge contour edge
•
轮廓提取及拟合
2.3 曲线优化 optimizing the curves 求解: relaxation procedure 局部最优
•
轮廓提取及拟合
2.3 防止过优化 corner detection
•
图像绘制
3.1 Rendering Truncated Gaussian influence function we place truncated Gaussian influence functions ( = 1, radius 2 pixels) at the cell centroids and set their support to zero outside the region visible from the cell centroid.
The final color at a point is computed as the weighted average of all pixel colors according to their respective influence.
4.1 计算速度 •
结论
4.2 实验结果 •
结论
see the paper