基于多特征融合的抽象画情感研究  

Research on emotion of abstract painting based on multi-feature fusion

作　　者：白茹意 郭小英 贾春花[1] Bai Ruyi;Guo Xiaoying;Jia Chunhua(School of Automation&Software,Shanxi University,Taiyuan 030013,China)

机构地区：[1]山西大学自动化软件学院,太原030013

出　　处：《计算机应用研究》2021年第2期587-590,共4页Application Research of Computers

基　　金：国家自然科学青年基金资助项目(61603228);山西省面上青年基金资助项目(201901D211171);山西省回国留学人员科研教研项目(HGKY2019001);山西省高等学校科研创新项目(2020L0036)。

摘　　要：抽象画作为一种寓意含蓄的艺术作品,传递出的情感也是含蓄的,确定其情感分类也比较困难,为此采用多特征融合方式预测抽象画的情感。首先采用K-means聚类提取抽象画图像的主色调作为底层颜色特征,采用灰度—梯度共生矩阵提取底层纹理特征,采用卷积神经网络自动提取高层语义特征;其次由于特征维度不同,采用多核学习对底层和高级语义特征进行融合;最后采用支持向量机实现抽象画情感识别,分为积极与消极两类。在MART数据集上进行测试,并与其他现有分类模型进行了比较,实验结果显示该方法在测试性能上优于已有模型。Abstract painting,as an implied work of art,conveys implicit emotions.Therefore,it is also difficult to determine the emotion of abstract painting.This paper used multiple feature fusion to predict the emotions of abstract painting.Firstly,it used K-means clustering to extract the main tone of abstract image as the underlying color feature,used gray-gradient co-occurrence matrix(GGCM)for the underlying texture feature,and used CNN network to extract the high-level semantic feature automatically.Secondly,because of the different feature dimensions,it adopted multiple kernel learning(MKL)to fuse the underlying and high-level semantic features.Finally,it applied MKL-SVM to realize the emotion recognition of abstract pain-ting,which included positive and negative.This paper tested on MART data set.The experimental results show that the proposed method outperforms the existing models in test performance.

关 键 词：情感识别 灰度—梯度共生矩阵 特征融合 多核学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]