基于多维特征分布的设计风格偏好度量模型研究  

Design style preference measurement model based on convolution neural network

作　　者：姜鑫玉[1] 栾宽 唐智[1] 王平 JIANG Xinyu;LUAN Kuan;TANG Zhi;WANG Ping(School of Mechanical Engineering,Donghua University,Shanghai 201620,China;College of Intelligent Systems Science and Engineering,Harbin Engineering University,Harbin,Heilongjiang 150001,China)

机构地区：[1]东华大学机械工程学院,上海201620 [2]哈尔滨工程大学智能科学与工程学院,黑龙江哈尔滨150001

出　　处：《燕山大学学报》2021年第2期174-179,188,共7页Journal of Yanshan University

基　　金：国家自然科学基金资助项目(51775106);上海市教委教育科学基金资助项目(C2-2020047)。

摘　　要：利用计算机对用户所需要的设计风格进行准确评估能够使设计变得高效。本研究提出一种针对设计风格偏好进行数字化评估的模型。模型采用卷积神经网络提取不同维度的图像特征,计算特征向量的格莱姆矩阵表示图像的特征分布,用图像特征分布间的均方误差作为描述图像风格的分值,通过比较不同图像的风格分值评价用户的设计风格偏好。该模型通过室内设计效果图进行了性能验证。结果表明该模型能够区分不同风格设计作品,同类风格设计作品排序与设计师的符合率达90%,可用于用户设计风格偏好的评估。An efficient design can be achieved by accurate evaluation of user′s design style based on computer.A digital evaluation model for design style preference is presented in this paper.A convolution neural network is used to extract image features of different dimensions,and the Gram matrix of the feature vector is calculated to represent the feature distribution of the image.The mean square error of the image feature distribution is used as the score to describe the image style,and the user′s design style preference is evaluated by comparing the score of different image styles.The performance of the model is verified by the interior design renderings.Different styles of design works can be distinguished by the model,and the coincidence rate between the model and designers is 90%in the ranking of similar styles of design works.These results suggested that the model can be used to evaluate the user′s design style preference.

关 键 词：设计风格 偏好 多维特征 特征分布 度量模型 

分 类 号：J045[艺术—艺术理论] TP391.7[自动化与计算机技术—计算机应用技术]