高级检索
李瑞瑞, 孙世豪, 杨孝辉, 胡伟. 基于多层次显著图的自动铅笔画生成[J]. 计算机辅助设计与图形学学报, 2018, 30(1): 107-115. DOI: 10.3724/SP.J.1089.2018.16161
引用本文: 李瑞瑞, 孙世豪, 杨孝辉, 胡伟. 基于多层次显著图的自动铅笔画生成[J]. 计算机辅助设计与图形学学报, 2018, 30(1): 107-115. DOI: 10.3724/SP.J.1089.2018.16161
Li Ruirui, Sun Shihao, Yang Xiaohui, Hu Wei. Automatic Generation of Pencil Drawings Based on Multi-level Saliency Maps[J]. Journal of Computer-Aided Design & Computer Graphics, 2018, 30(1): 107-115. DOI: 10.3724/SP.J.1089.2018.16161
Citation: Li Ruirui, Sun Shihao, Yang Xiaohui, Hu Wei. Automatic Generation of Pencil Drawings Based on Multi-level Saliency Maps[J]. Journal of Computer-Aided Design & Computer Graphics, 2018, 30(1): 107-115. DOI: 10.3724/SP.J.1089.2018.16161

基于多层次显著图的自动铅笔画生成

Automatic Generation of Pencil Drawings Based on Multi-level Saliency Maps

  • 摘要: 图像艺术风格化是将图像自动转化为具有艺术风格的图像的方法统称,具有十分广阔的需求和市场应用前景.本文针对现有方法生成的图像往往风格单一、线条僵硬,提出了一种基于多层次视觉显著图将照片图像自动转化成铅笔画的方法,借助视觉显著性特点,使得风格化图像更加自然逼真.首先利用量子切割原理生成一幅包含多个层次的视觉显著性映射图,为每个像素计算一个重要性概率值;其次,利用显著性映射图对铅笔画轮廓图进行建模,并利用图像的边缘切线流为边缘像素计算显著性层次,生成不同粗细、长短、方向的笔画;最后,利用显著性映射图对色调纹理进行建模,对不同层次的像素采用不同程度的灰度变换和纹理合成.与现有的非真实感图像生成方法相比,文中方法由于利用了多层次视觉显著性信息,生成的图像更加具有层次感和艺术表现力,更接近画家绘画作品.实验表明,采用该方法得到的合成框架可以将照片快速转换成铅笔画艺术风格图像,效果生动逼真.

     

    Abstract: Artistic stylization is a kind of techniques that could automatically convert photos to art-style images.It has a board market and a vast potential for future development.Currently,images generated by the existing artistic stylization methods are often of rigid style with stiff strokes.This paper proposed a non-photo realistic image synthesis framework for pencil drawing generation.Based on the multi-level saliency map,the generated pencil drawings are more natural.The method first generated a multi-level saliency map through quantum cut algorithm,which computed a salience probability for each pixel in the photo;then it built models to synthesize sketches,which created strokes of various length,thickness,and orientation according to the saliency level by using the edge tangent flow algorithm;finally,it built models to synthesize tones,which transferred the brightness in gray-scale and performed texture mapping in the level of details.Compared with the state-of-the-art non-photo realistic rendering methods,our method could get more layered and vivid pictures which are closer to real artistic works.Experiments show that the synthesis framework transfer photos to pencil drawings fast and the transferred images have vivid artistic effects.

     

/

返回文章
返回