计算美学:计算科学驱动的视觉美学度量与生成

目的 计算美学旨在实现可计算的审美决策、美学设计与艺术创作,其在美学的计算建模与高效分析理解方面具有重要的科学意义.同时,依赖于计算机技术自动、高效等特点,以计算方法改进美学工具、进行艺术作品创作具有广阔的应用前景.随着人工智能技术的发展,计算美学的内涵得到了极大的丰富与扩展,同时也出现了对美学更高层次的理解、创作的挑战.方法 从现有前沿工作出发,将计算美学研究归纳为美学度量与美学生成两个方面研究内容.其中美学度量通过神经网络自动提取图像美学特征来判断视觉作品是否符合美感,美学生成则主要通过生成模型自动进行图像风格、布局、颜色等的设计.结论 通过分析计算美学发展的关键挑战对审美认知机理驱动的美学度量、个性化美学生成等未来方向进行了展望.

Computational Aesthetics:Visual Aesthetics Measurement and Generation Driven by Computational Science

Computational Aesthetics aims to achieve computable aesthetic decision-making, aesthetic designing and artistic creating, and it has important significance in the computational modeling of aesthetics and efficient understanding of the aesthetic. At the same time, based on the automatic and high-efficiency characteristics of the computer technology, it has promising application prospects to utilize computational methods to improve aesthetic tools and create artistic works. With the gradual maturity of AI, the connotation of computational aesthetics has been greatly enriched and expanded, and meanwhile, there has been challenges to higher-level understanding and creation of aesthetics. Starting from current frontier works, this paper summarizes the research of the Computational Aesthetics into two aspects:aesthetic measurement and aesthetic generation. The aesthetic measurement uses neural networks to automatically extract the aesthetic features to judge whether the visual works meet the aesthetic feeling, and the aesthetic generation mainly utilizes the generation models to automatically design the styles or colors of target images. Finally, this article analyzes the key challenges in the development of computational aesthetics, and discusses the future directions such as the cognition mechanism-based aesthetic measurement, and the personalized aesthetics generation.