人脸肖像风格化绘制技术研究
发布时间：2018-11-03 21:47
【摘要】：非真实感绘制(Non-Photorealistic Rendering,简称NPR)是计算机技术与绘画艺术相结合的一个研究领域,其研究的主要是用计算机软硬件手段模拟具有不同绘画风格的艺术流派,如铅笔画、水彩画、油画等。其研究成果可用于广告宣传、室内装修、影视制作、卡通画制作等许多领域。 人脸肖像风格化绘制是非真实感绘制研究中的一个分支,本文对肖像风格化绘制研究的背景、意义及其中用到的一些关键技术展开研究,并给出相应的实验结果效果图。 首先,论文选取丁绍光为代表的云南重彩画作为研究对象,提出云南重彩画风格的人脸肖像生成算法,对人脸照片进行重彩画风格的转换。该算法作为基于照片的云南重彩画风格化绘制研究中的一个重要部分,其结果将会影响到整体风格的绘制效果。云南重彩画人脸肖像生成算法,探究并总结了云南重彩画人脸绘制规律,采用移动最小二乘法实现云南重彩画风格化转换中五官特征的变形,采用基于流的高斯差分滤波器(Flow-based Difference-of-Gaussians,简称FDOG),进行人脸肖像轮廓线的提取。此算法能够有效地模拟云南重彩画的人脸绘制特点,并取得了较好的绘制效果。 其次,对卡通风格的肖像绘制算法进行了研究。针对卡通画特点,我们先对输入图像采用双边滤波进行抽象化处理,使得高对比度区域的对比度更高,低对比度区域的对比度更低,简化了场景信息。然后,对抽象化图像进行量化处理以模拟卡通画的着色风格,并采用FDOG技术进行边缘轮廓线的提取以提高卡通画的可视性；最后将量化后的图像与提取出的边缘图进行融合,即可得到色彩鲜艳、可视性较高、细节较少的卡通画风格的效果图。 最后,针对图像量化处理过程中量化区间之间出现的明显过渡不连续问题,论文中采用颜色软量子化方法改进量化效果。引入亮度梯度函数,这样,只有在亮度梯度大的地方,才会出现明显的边界,在梯度低的区域,边界被延展到较大的区域中；最后可得到色彩鲜艳、量化区间过渡连续的量化效果图。
[Abstract]:Non-Photorealistic Rendering, rendering (NPR for short) is a research field which combines computer technology with painting art. Its research mainly uses computer software and hardware to simulate art schools with different painting styles, such as pencil-drawing. Watercolor painting, oil painting, etc Its research results can be used in advertising, interior decoration, film and television production, cartoon production and many other fields. Facial portrait stylization is a branch of non-realistic rendering. In this paper, the background, significance and some key techniques of portrait stylization rendering are studied, and the corresponding experimental results are given. Firstly, the thesis chooses Yunnan heavy color painting represented by Ding Shaoguang as the research object, and puts forward a face portrait generation algorithm of Yunnan heavy color painting style. As an important part of the research on the stylized rendering of Yunnan heavy color paintings based on photos, the result of the algorithm will affect the rendering effect of the whole style. Yunnan heavy color painting face portrait generation algorithm, the study and summary of Yunnan heavy color painting face rendering law, using the moving least square method to realize the transformation of Yunnan heavy color painting style transformation of the features of the five features, A stream-based Gao Si differential filter (Flow-based Difference-of-Gaussians, for short FDOG),) is used to extract face profile. This algorithm can effectively simulate the features of Yunnan heavy color painting face rendering, and achieve a good result. Secondly, the image rendering algorithm of cartoon style is studied. According to the characteristics of cartoon, we first use bilateral filtering to abstract the input image, which makes the contrast of high contrast region higher and low contrast region lower, which simplifies the scene information. Then, the abstract image is quantitatively processed to simulate the coloring style of the cartoon, and the edge contour is extracted by using FDOG technology to improve the visibility of the cartoon. Finally, by merging the quantized image with the extracted edge image, we can get the cartoon drawing with bright color, high visibility and less detail. Finally, aiming at the obvious transition discontinuity between the quantization intervals in the image quantization process, the color soft quantization method is used to improve the quantization effect in this paper. The luminance gradient function is introduced so that only where the luminance gradient is large can the obvious boundary appear. In the region with low gradient the boundary is extended to the larger region. Finally, the quantization effect diagram with bright color and continuous transition of quantization interval can be obtained.
【学位授予单位】：云南大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.41
本文编号：2309133
[Abstract]:Non-Photorealistic Rendering, rendering (NPR for short) is a research field which combines computer technology with painting art. Its research mainly uses computer software and hardware to simulate art schools with different painting styles, such as pencil-drawing. Watercolor painting, oil painting, etc Its research results can be used in advertising, interior decoration, film and television production, cartoon production and many other fields. Facial portrait stylization is a branch of non-realistic rendering. In this paper, the background, significance and some key techniques of portrait stylization rendering are studied, and the corresponding experimental results are given. Firstly, the thesis chooses Yunnan heavy color painting represented by Ding Shaoguang as the research object, and puts forward a face portrait generation algorithm of Yunnan heavy color painting style. As an important part of the research on the stylized rendering of Yunnan heavy color paintings based on photos, the result of the algorithm will affect the rendering effect of the whole style. Yunnan heavy color painting face portrait generation algorithm, the study and summary of Yunnan heavy color painting face rendering law, using the moving least square method to realize the transformation of Yunnan heavy color painting style transformation of the features of the five features, A stream-based Gao Si differential filter (Flow-based Difference-of-Gaussians, for short FDOG),) is used to extract face profile. This algorithm can effectively simulate the features of Yunnan heavy color painting face rendering, and achieve a good result. Secondly, the image rendering algorithm of cartoon style is studied. According to the characteristics of cartoon, we first use bilateral filtering to abstract the input image, which makes the contrast of high contrast region higher and low contrast region lower, which simplifies the scene information. Then, the abstract image is quantitatively processed to simulate the coloring style of the cartoon, and the edge contour is extracted by using FDOG technology to improve the visibility of the cartoon. Finally, by merging the quantized image with the extracted edge image, we can get the cartoon drawing with bright color, high visibility and less detail. Finally, aiming at the obvious transition discontinuity between the quantization intervals in the image quantization process, the color soft quantization method is used to improve the quantization effect in this paper. The luminance gradient function is introduced so that only where the luminance gradient is large can the obvious boundary appear. In the region with low gradient the boundary is extended to the larger region. Finally, the quantization effect diagram with bright color and continuous transition of quantization interval can be obtained.
【学位授予单位】：云南大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.41
【参考文献】
中国期刊全文数据库 前10条
1 钱小燕;肖亮;吴慧中;;一种流体艺术风格的自适应LIC绘制方法[J];计算机研究与发展;2007年09期
2 邓维;魏迎梅;张友根;;利用图像变形生成个性化人脸卡通[J];计算机工程与应用;2011年24期
3 李龙生;周经野;陈益强;刘军发;陈振宇;;面向移动的卡通人脸生成系统[J];计算机科学;2007年01期
4 周仁琴;周经野;陈益强;刘军发;;基于特征发现的卡通人脸肖像生成[J];计算机辅助设计与图形学学报;2006年09期
5 阎芳;费广正;柳婷婷;马文慧;石民勇;;漫画风格的人脸肖像生成算法[J];计算机辅助设计与图形学学报;2007年04期
6 华顺刚;刘婷;;利用移动最小二乘法实现图像变形[J];计算机应用;2009年01期
7 陈文娟;石民勇;孙庆杰;;人脸漫画特征关系夸张及相像度优化[J];计算机应用;2009年S2期
8 万里红;董兰芳;沈育平;陈意云;;人物肖像画及其卡通动画仿真系统的建立[J];计算机仿真;2007年05期
9 赵杨;徐丹;;运用流体模拟的油画生成方法[J];软件学报;2006年07期
10 普园媛;苏迤;魏小敏;尉洪;钱文华;徐丹;;云南重彩画白描图绘制系统[J];计算机工程与设计;2011年02期
中国硕士学位论文全文数据库 前1条
1 覃日钊;线积分卷积与纹理合成在云南重彩画风格化绘制上的应用[D];云南大学;2011年
，本文编号：2309133
本文链接：https://www.wllwen.com/wenyilunwen/guanggaoshejilunwen/2309133.html
