Selective image abstraction
Selective image abstraction
复制标题
DOI：
10.1007/s00371-010-0522-2
复制
发表时间：
2011-03
期刊：
影响因子：
--
通讯作者：
Lin Cong;Ruofeng Tong;Jinxiang Dong
中科院分区：
文献类型：
--
作者：
Lin Cong;Ruofeng Tong;Jinxiang Dong
我们提出了一种新的、方便的方法来产生有选择的图像风格化简化。用户使用画笔交互地标记输入图像的某些区域，这些区域将保持不变。然后将这些区域的边界自动优化为图像中的底层对象边界。然后，我们的方法对未标记区域执行程式化简化，同时保留标记区域。该方法确保了风格化和未改变区域之间的平滑过渡，留下了真实和抽象相结合的混合现实图像。程式化简化使用非线性扩散，可以产生复杂的结果。我们修改了经典的非线性扩散模型，加入了双边滤波；我们根据用户的输入对每个像素进行扩散速度控制。根据扩散时间可以直观地控制简化程度；另一个参数控制抽象样式，提供简单直观的用户界面。我们的贡献包括一种简单易用的方法来产生一种新的NPR风格和一个改进的非线性扩散模型，适合于这种选择的风格化简化任务。实验结果表明，最终的混合现实结果是和谐的。
We present a novel and convenient method for producing selective stylized simplification of images. The user uses a brush to interactively mark certain areas of the input image which are to be left unaltered. Boundaries of these areas are then automatically optimized to underlying object boundaries in the image. Our method then performs stylized simplification of the unmarked areas, while preserving the marked areas. The method ensures a smooth transition between stylized and unaltered regions to leave amixed realityimage which combines the real and the abstract. Stylized simplification is performed using nonlinear diffusion, which can generate sophisticated results. We modify the classic model of nonlinear diffusion to incorporate bilateral filtering; we apply diffusion speed control of each pixel based on the user’s input. The level of simplification can be controlled intuitively based on the diffusion time; another parameter controls the abstraction style, giving a simple and intuitive user interface. Our contributions include a simple-to-use method to produce a novel NPR style and a modified nonlinear diffusion model suited to this selective stylized simplification task. Experimental results show that the final mixed reality results are harmonious.