基于艺术风格相似性规则的绘画图像分类  

Painting image classification based on aesthetic style similarity rule

作　　者：杨冰[1] 许端清[1] 杨鑫[1] 赵磊[1] 唐大伟[1] 

机构地区：[1]浙江大学计算机学院网络与媒体实验室,310027

出　　处：《浙江大学学报（工学版）》2013年第8期1486-1492,共7页Journal of Zhejiang University：Engineering Science

基　　金：数字博物馆核心关键技术攻关资助项目(2012BAH03F02);国家"973"重点基础研究发展计划资助项目(2012CB725305)

摘　　要：针对传统分类方法中直接建立描述符描述绘画的艺术风格而忽略艺术风格这一对象本身所具有的属性的问题,从人脑认知机制出发,将艺术风格的描述建立于特征向量及相似性原理的基础之上.算法基于相似性原理,通过挖掘艺术风格自身属性,建立艺术风格相似性规则.在遵守相似性规则的基础上,对艺术领域中普遍认可的风格特征进行量化,建立图像的自相似性描述符,采用距离函数计算出图像与其他所有样本图像之间的相似性系数,将由相似性系数构成的相似性矩阵作为该图像的风格相似性矩阵,采用Adaboost分类器对未知样本做出判断.结果表明艺术风格相似性规则的有效性,在不同艺术风格绘画图像上的分类实验也得到较好的结果.For the problem of the traditional classification methods that often directly construct descriptors and ignoring the inherent properties of aesthetic style,starting from human brain cognitive mechanisms,we propose to describe the aesthetic style on the basis of the feature vector and similarity principle.We create aesthetic style similarity rule（ASSR） based on similarity principle and deep analysis of the inherent properties of aesthetic style.Follow ASSR,we then quantify style features that are generally accepted in realm of art to build the image self-similarity descriptor.The distance function is used in our method to compute the similarity coefficient between one image to all other images.Finally,the similarity matrix composed of similarity coefficients could be treated as the aesthetic style similarity descriptor for the image,and Adaboost classifier is taken to evaluate the unknown images.The results demonstrate the efficiency of ASSR and our method for the painting image classification with different aesthetic style obtains good performance.

关 键 词：艺术风格 相似性规则 图像分类 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]