李炳彤 张华熊 康锋 林翔宇
摘要:针对分形服饰图案的特点,提出了一种多特征融合的美学评价方法。首先对设计图像进行主观美感评分,建立样本图像库;然后提取图像的美学复杂度、色彩分布、构图和纹理等19维美学特征以及7维通用图像特征,建立特征库;通过支持向量机建立图像美感分类器,最终实现高、低美感图案的自动分类。实验结果显示:该方法的高、低美感评价准确率为87.14%,AUC值为0.923 7,符合人类对图案的审美思维;与典型的美学分类算法对比,该方法对分形图案美感等级分类的性能有所提高。
关键词:服饰图案;美学评价;分形图案;美学特征;支持向量机
中图分类号:TN919;TS145.4文献标志码:A文章编号:1009-265X(2018)03-0047-06Aesthetic Assessment of Fractal Dress Pattern Based on MultiFeature Fusion
LI Bingtong, ZHANG Huaxiong, KANG Feng, LIN Xiangyu
(School of Information and Technology, Zhejiang SciTech University, Hangzhou 310018, China)Abstract:For features of fractal dress pattern, this paper proposes an aesthetic evaluation method based on multifeature fusion. Firstly, each image was graded according to the subjective aesthetic perception and an image database was established. Then, 19 dimensional aesthetic features and 7 common image features were extracted to establish feature database, including aesthetic complexity, color distribution, composition and texture, etc. and combine wit. An image aesthetic classifier was created through SVM to achieve automatic classification of high and low aesthetic patterns. The experiment results demonstrate that the accuracy of high and low aesthetic evaluation is 87.14%, and the AUC value is 0.923 7. The results show that the method proposed in this paper is according with human aesthetic thinking. Compared with typical algorithms of aesthetic classification, this method improves the performance of aesthetic grade classification for fractal patterns.
Key words:dress pattern; aesthetic evaluation; fractal pattern; aesthetic feature; SVM
通信作者:張华熊,Email:zhxhz@zstu.edu.cn计算机可以通过一定的数值计算方法生成变化无穷、玄妙美丽的分形图案,这些图案具有局部和整体的自相似性、动态平衡性、对称性以及精细的结构,内涵丰富,风格独特,具有科学和艺术的双重价值。分形图案在中国服装纺织业已被广泛应用于服饰图案设计,并得到了很多业内外人士的肯定[15]。然而,这些分形图案是数学的产物,由纯粹的公式生成,在一定程度上存在着机械性和随机性。设计师能够在计算机生成的分形图像中挑选出可以表达设计思想和创作灵魂的图案,但是大部分分形图像不能自动地符合设计师的主观思想和用户的感受。因此,采用数字图像处理技术,用计算机对分形服饰设计图案进行客观的美学评价,不仅可以方便设计师挑选合适的服饰图案,而且对服饰图案的设计起到指导作用,激发设计师的创作灵感。……