基于深度学习的界面设计美学评价研究

Aesthetic Evaluation of Interface Design Based on Deep Learning

中文关键词: 深度学习  卷积神经网络  闪屏图像  美学质量评价  美学评价分布

英文关键词: deep learning  convolutional neural network  launch screen image  aesthetic quality evaluation  aesthetic evaluation distribution

中文摘要:

      目的 美感已经成为人机交互（HCI）的核心结构之一，对用户的感知和态度具有明显的有益影响。然而界面美观性评价方法仍是设计师及其团队所面临的重要问题。引入深度学习技术来探讨其评价界面设计美感的可能性。方法 分别使用基于深度卷积神经网络的闪屏美学分类方法和Google提出的基于深度学习NIMA神经网络，来预测闪屏图像的美学评价分布。结果 通过研究发现，使用基于深度学习NIMA神经网络可以得到比传统方法更具体的评价结果，帮助设计师有效而客观地评价界面设计。结论 将计算机图像美学评价的研究领域拓展到界面设计领域，验证了深度卷积神经网络在界面设计美学评价领域使用的可行性。未来图像美学评价还可以介入更多的设计相关领域，辅助设计师做出更有效的设计和商业决策。

英文摘要:

      Aesthetics has become one of the core structures of human-computer interaction (HCI), which has a significant beneficial effect on user perception and attitude. However, the evaluation method of interface aesthetics is still an important issue faced bydesigners and their teams. The work aims to explore the possibility of evaluating the aesthetics of interface design with deep learning techniques. The aesthetic classification method of launch screen based on deep convolutional neural network andthe neural network based on deep learning NIMA proposed by Google were used to predict the aesthetic evaluation distribution of launch screen images. Through research, it was found that theneural network based on deep learning NIMA couldbe used to get more specific evaluation results than traditional methods, which couldhelp designers to evaluate interface design effectively and objectively. The research field of computer image aesthetic evaluation is expanded to the field of interface design, which verifies the feasibility of using deep convolutional neural network in the field of aesthetic evaluation oninterface design. In the future, image aesthetic evaluation can also intervene in more design-related fields to assist designers in making more effective design and business decisions.